* According to the RFC[1], a server may ignore a Range header and
   should ignore a Range header containing units it doesn't
   understand.  WB was erroring under these conditions, but it seems
   more appropriate to ignore the field.  Testing against external
   providers showed no consistent practice to follow.  The
   parse_request_range docs have been updated to reflect the new
   behavior.  Big thanks to @birdbrained for doing the legwork on
   researching this!

 [1] https://tools.ietf.org/html/rfc7233#section-3.1

This is a rebased and squashed commit of a fully working
implementation for the provider based on Google Cloud
Storage's JSON API and OAuth 2.0 protocol.

For detailed commits and messages, please see this PR:
CenterForOpenScience#317

- Removed functionalities that are not used by
  OSFStorage and that do not have documented
  support
- Removed JSON API related docstrings/comments
- Added TODOs on what needs to be refactored
- Update tests

- updated settings to use XML API
- updated metadata.py to parse response headers
  and init metadata object
- added a helper function in utils.py to convert
  GCS's base64-encoded hash to hex digest
- refactor the structure of fixtures and updated
  them with real responses from Postman tests

- Fully refactored all provider actions to use
  XML API and signed request and implemented a
  minimal version:
    - Upload
    - Download
    - Metadata for file
    - Delete file
    - Intra-copy file
- Added TODO comments for Phase 1, 1.5 and 2:
    - Create folder
    - Metadata for folder
    - Delete folder
    - Intra-copy folder
- Rewrote the provider's `.build_url` with
  `.build_and_sign_req_url` to take care of the
  url buidling and request signing together

Discovered and fixed a few issues during OSF integration
test and updated comments and docstr.

- Main issue: aiohttp parses the `x-goog-hash` correctly
  and returns an `aiohttp.MultiDict` dictionary that
  contains two entries with the same key, one for crc32c
  and one for md5. Fix: updated headers parsing and tests
- Minor fixes/updates
  - Updated HTTP method for delete and intra-copy
  - Added bucket to "x-goog-copy-source" and no longer
    convert value to lower case for request signing
  - Strip '"' from "ETag" for verifying upload checksum
  - Removed "Content-length" and only use "x-goog-stored-
    content-length" for file size
  - Prefixed `build_and_sign_req_url()` with `_`

Note: environment variables are passed in as string

 Some tests and refactors to come, but provider is ready for testing.

 [SVCS-617]
 Closes: CenterForOpenScience#322

- WB only needs to know the bucket name, OSF
  handles the region and select the bucket with
  the expected region for WB.

[SVCS-617]
Closes: CenterForOpenScience#329

- Removed deprecated `fixture.py`, which is now
  replaced by `providers.py`, `folders.py` and
  `files.py` in the `fixtures/` directory.
- Updated fixutres for CRUD operations and added
  back CRUD tests that were accidentally removed
- Fix exipration check in utility tests where it
  now use the `settings.SIGNATURE_EXPIRATION` to
  calculate the expected exipiration time.

Remove redundant type casting for expiration

- Type casting is now done in the settings.py. I
  assume that this piece of code is accidentally
  left here during the last merge.

- Fix imports order. `typing` is a standard lib
- Use `+=` for request segments contatenation
- Replace `return True if <condition> else False`
  with `return bool(<condition>)` where the
  condition is `None` or `not None`

- Update both the metadata method and the utility
  function to use the strict regex matching based
  on the RFC specification for Base 64 encoded
  crc32c and md5 hashes
- Google Cloud uses the standard alphabets for
  Base 64 encoding: [A-Za-z0-9+/=]
- RFC reference:
  http://www.rfc-editor.org/rfc/rfc4648.txt

- Added an alternative constructor for the base
  metadata which takes either an standard python
  dict or a pair of object name and multi-value
  dict during initialization
- Updated its usage in the provider
- Changed @staticmethod to @classmethod for
  `get_metadata_from_resp_headers()`
- Added metadata tests for both successful and
  failed initialiaztion.

- Added `get_multi_dict_from_json()` (and a test
  for it) to utils so that all tests now use this
  helper method to build resp headers
- Refactored the metadata structure to test three
  classes separately
- Removed import alias such as `core_exception`,
  `pd_settings` and `pd_utils` since there is no
  shadowing issues any more

- Both GC file and folder metadata now uses this
  decidated method to initialize with object name
  and aiohttp's "MultiDict" response headers
- Removed the alternative constructor for GC base
  metadata and updated related code and tests

- Moved quirks in comments into DocStr so that
  that they are available in WB Docs
- Use `` (double tildes) for code in DocStr
- Fixed Sphinx warnings

Side effect: modified the function signature for
get_multi_dict_from_python_dict() to expect dict
instead of json; updated all related tests.

- GC's private memebers are now availabe in the
  WB Docs by adding :private-members:
- Remove :inherited-members: and :undoc-members:
- Updated the .rst files to include GC, metadata
  and utils

- mppy doesn't handle class inheritence well
- mypy has a problem with **{} arguments
- mypy doesn't handle multiple options of
  return type well

- Upload fails due to CIMultiDcitProxy inherits
  from MultiDictProxy but not from MultiDict.
  However, aiohttpretty returns CIMultiDict while
  aiohttp returns CIMultiDictProxy, the type is
  strict and only recognize CIMultiDict
- Updated the check statement with the either
  MultiDict or MultiDictProxy:
  - WB code uses CIMultiDcitProxy (a subclass of
  MultiDictProxy) since aiohttp parses the hash
  headers already.
  - WB test code uses MultiDict to modify the
  dictionary in get_multi_dict_from_python_dict()
  which returns MultiDcitProxy.

 * Minor updates to formatting of method sigantures, import order, and
   judicious application of DeMorgan's Law to make conditionals more
   readable.

 * Update formatting of docstrings to make Sphinx docs more
   cross-linked and browsable.

 * Minor style fixes.

 * Update formatting of docstrings to make Sphinx docs more
   cross-linked and browsable.

 * Minor updates to formatting of method sigantures.

 * Update formatting of docstrings to make Sphinx docs more
   cross-linked and browsable.

 * Minor style fixes for import order and signature formatting.

 * Remove unused fixtures and imports from tests.

 * __init__.py doesn't need to export the Metadata classes.  Remove
   that and update the test files that were using it.

 Code improvements, tests, minor fixes for the limited Google Cloud
 Storage provider.

 [SVCS-617]
 Closes: CenterForOpenScience#327

- The type fix also finds a bug in our code where
  `._intra_move_copy_metadata()` calls a buggy
  `._get_folder_meta()` that further calls
  ._serialize_item() with invalid arguments.

[SVCS-595]
Closes: CenterForOpenScience#332

 * ReadTheDocs has updated their base python image, meaning the
   anaconda-based config is no longer needed.

   The conda config was out-of-date and failing to build anyway since
   the setuptools dependency version bump.

 * Stop sending download callbacks for 206 Partial responses.  These
   should not be counted as full downloads.  WB does not directly
   support Range requests on direct-from-provider downloads (signed
   urls), but at least curl and Postman appear to propagate Range
   headers from the original request to the follow-up redirection
   request.  For now, log 302 responses with Range headers, but
   continue to send download callbacks as normal.  The logs will be
   used to determine the correct behavior in the future.

 Closes: CenterForOpenScience#336

 * Turn off revisions metadata logging.  When file download logging
   was added regular metadata requests were excluded, but revisions
   were overlooked.

 * Update WB to return the request method, url, user agent, and
   referrer url in the logging callback payload.  Intended to help the
   callback listener provide more specific download metrics.

 [SVCS-673]
 Closes: CenterForOpenScience#324

 * Otherwise an "unclosed response" error will appear when the next
   celery task is scheduled.

 * The osfstorage provider kicks off two tasks after upload; one to
   backup the file to Amazon Glacier and one to generate parity files
   that are sent to a bucket on the storage backend provider.  Since
   both tasks run in parallel and need a copy of the uploaded file to
   work, neither could be responsible for deleting it when done.
   Instead this work would need to be done periodically by an admin to
   keep the disk from filling up.

   Both tasks are now included in a Celery chord.  When all the tasks
   in a chord are done, it runs another task.  In this case, the chord
   will run a cleanup task after both other tasks finish.  To make
   this simpler, each upload is moved to a temporary directory where
   it and its generated parity files lived.  This temporary directory
   is removed by the cleanup task.

   The parity and archive tasks tests have been commented out rather
   than updated, since a simpler approach may be implemented soon.

- Use `functool.partial()` to delay building and
  signing URL unitl the request is actually made.
- Now `make_request()` get a brand new URL every
  time it retries a failed request.

 [SVCS-697]
 Closes: CenterForOpenScience#339

 * Otherwise an "unclosed response" error will appear when the next
   celery task is scheduled.

 * The osfstorage provider kicks off two tasks after upload; one to
   backup the file to Amazon Glacier and one to generate parity files
   that are sent to a bucket on the storage backend provider.  Since
   both tasks run in parallel and need a copy of the uploaded file to
   work, neither could be responsible for deleting it when done.
   Instead this work would need to be done periodically by an admin to
   keep the disk from filling up.

   Both tasks are now included in a Celery chord.  When all the tasks
   in a chord are done, it runs another task.  In this case, the chord
   will run a cleanup task after both other tasks finish.  To make
   this simpler, each upload is moved to a temporary directory where
   it and its generated parity files lived.  This temporary directory
   is removed by the cleanup task.

   The parity and archive tasks tests have been commented out rather
   than updated, since a simpler approach may be implemented soon.

 * These commits were originally merged to develop, but are being
   hotfixed into master to solve issues with unbounded storage
   consumption.

 * In 0.38.2, the googlecloud provider was updated to provide a
   function as the url parameter to `BaseProvider.make_request`.
   Since googlecloud urls are signed and can expire, it must delay
   generation until right before issuing.  Otherwise, if the first
   request fails, the second may not get issued until after the
   signature has expired.

   Unfortunately, `.make_request` was evoking the url function outside
   of the retry loop.  This resulted in the same signed url being used
   for each retry.  Moving this inside the retry loop will cause new
   urls to be generated for each retry request.

 Closes: CenterForOpenScience#340

 [#SVCS-689]

 * osfstorage is being updated to include a flag in file metadata that
   will indicate if the requesting user has seen the most recent
   version of the file.  To help the OSF properly determine this,
   update WB to send along the requesting user's id when asking for
   the metadata of all files in a given directory.

   The requesting user is not the same as the authorizing user.  If
   Barbara asks for the contents of an osfstorage directory created by
   Alice, Barbara is the *requesting* user, while Alice is the
   *authorizing* user.  WB first verifies that Barbara has the
   necessary access to the file, but uses a shared secret to retrieve
   metadata about the file.  This change is necessary to inform the
   OSF who is behind the request.

 * The osfstorage folder children response is updated to include a new
   flag, `latestVersionSeen`.  If this flag is `null`, the requesting
   user has never seen *any* version of the file.  If it is `true`,
   the user has seen the latest version of the file.  If it is
   `false`, the user has seen a previous version of the file, but not
   the most recent one.  This flag will be exposed through the
   `extra.latestVersionSeen` flag in OsfStorageFileMetadata.

 * Due to a quirk in WB and the OSF, the latestVersionSeen flag will
   only be correctly set on the responses from folder metadata list
   requests.  Neither service correctly handles requests for
   previous-version metadata for individual files.

 * Update tests to include latestVersionSeen in children response.

 [SVCS-689]
 Closes: CenterForOpenScience#337

 Closes: CenterForOpenScience#342

- This only applies to `exception_from_response()`
- Side-effect: fix not-released response

 [SVCS-693]
 Closes: CenterForOpenScience#338