WIP: Deprecating access_ssdl argument

[perolavsvendsen]

Ref discussion related to #540 and other issues

Proposing to deprecate the access_ssdl input argument in favor of classification and rep_include.

This PR does:

• Deprecate access_ssdl argument, introduces classification and rep_include arguments
• This is a follow-up from non-merged PR Populate the access.classification field #320 which was only partially made
• Stops modifying the config during runtime. Read it, then use the elements from it further.
• Moves some of the logic related to the config from dataio.py to global_configuration.py.

Might not be merged as-is, but perhaps elements or principles from it will be useful.

Purpose:

• Make the classification argument a bit more intuitive and readable
• Remove impression that this is something specific to SSDL when it is in fact not.

History:
Initially, the access.ssdl block was specific to SSDL and inherited from early versions of metadata when SSDL defined the FMU metadata (largely). Then, the access.ssdl.access_level was used also for access control outside SSDL (i.e. Sumo). Then, we decided to deprecate it - but only go half-way there, because of multiple challenges. Essentially, it stopped after adding access.classification to outgoing metadata (temporarily synchronized with access.ssdl.access_level.

The (milestone) goal should be to remove access.ssdl.access_level completely in favor of access.classification and then us this also in the config.

In short, we want to go from this:

exp = ExportData(access_ssdl={"access_level": "internal", rep_include=False})

...to this:

exp = ExportData(classification="internal", rep_include=False)

And we want to do the same in the config, i.e. from:

access:
  ssdl:
    access_level: internal
    rep_include: True

to:

access:
  classification: internal
  ssdl:
    rep_include: True

Complicating factors:

• We still need the rep_include parameter, so the access.ssdl block will remain for now
• We allow missing or non-valid config, and must handle that throughout fmu-dataio. This complicates things.
• We make changes to the config during runtime, which complicates things. (Changed in this PR.)
• We allow arguments to be given both to the initialization of ExportData and to the call to .generate_metadata(). This makes it a bit tricky to keep things in synch, particularly when we change things as we check them. (And then check them again later.)

Todo:

• Deprecate access_ssdl from input argument and from config (not discontinue it)
• Make support for using access.classification also in the config (and test it).
• Rewrite tests to primarily use the updated pattern (classification)
• Stop altering the config during runtime. Parse it, check it, then just read it.
• Discussion, and some more discussion

⚠️ This is work in progress, and not ready for merging. Perhaps it will never be.

[perolavsvendsen]

Had to include this to make warnings actually show, not sure if this is specific for me 🤷‍♂️

[perolavsvendsen]

Later learned that we should use other types of warnings to make sure they are actually seen by end users.

[perolavsvendsen]

Will remove this, but keeping it for now as a TODO: We need to make sure warnings are actually seen by end users/modellers. Possibly separate PR.

[perolavsvendsen]

Just sketching at this point, but some sort of pattern is perhaps starting to emerge

[perolavsvendsen]

Deprecating access_ssdl, adding classification and rep_include

[perolavsvendsen]

This is the deprecation warning intended for users who still use the access_ssdl argument

[perolavsvendsen]

The patterns below are the once we want to promote going forward

[perolavsvendsen]

ref discussion @tnatt
Possible "carve-out" from this:

• Better handling of config.
• Explicitly set the keys we need from config into class variables (Pydantic objects??), then not carry config further.

[perolavsvendsen]

@tnatt: rep_include does not have to be in the config, it will always be false as default anyway. I agree.

[perolavsvendsen]

Replaced by #628