build(devcontainer): 🧑‍💻 add support for reproducible env

[CandiedCode]

This PR is a continuation of #127 (comment)

Because a lot of this is json, this looks like a lot of lines of change. To keep this PR from growing larger, I only focused on 2 notebooks, XGBoost and pytorch. The other two can be updated to include tests, after this PR has landed.

• create a dev container with inital setup for python/poetry/pre-commit
• mount aws/secrets file that is expected in the pytorch notebook
  Note: I used the same key and secret AWS uses in their example seen here

$ export AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
$ export AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
$ export AWS_DEFAULT_REGION=us-west-2

• support black for notebooks and format the notebooks
• add nbmake tests to pytorch notebook and XGBoostNotebook
• add github action to run nbmake tests via devcontainer setup

---

The nbmake tests are on the metadata of the cells. It's not visible in the pull request view from github unless you look at the raw json of the notebook or review the PR in vscode with the gh pull request extension.

Example of reviewing PR with github pull request vscode extension

Example of nbmake tests in notebooks

Example where of a failing test to verify nbmake was working

Test model file exists

Test model is working as expected

Test model scan gives expected results by creating json report and expecting the values

Test unsafe model report contains expected issues

---

Note

While writing tests for the pytorch sentiment notebook and comparing the results of the scan from the console vs the results from the json, the console showed skipped.

Should this information not also be part of the summary in the json report?

Also if you like the Test / test-notebooks, since it also acts as end to end tests running model scan, I'd recommend that this is also part of the required actions.