Add workflow to build ml-metadata

[aktech]

This PR creates an action to build ml-metadata package via docker.

• Add build from docker
• Upload to PyPi
• Run tests via Pytest

Passing here on my fork: https://github.com/aktech/ml-metadata/actions/runs/11081764726/job/30793914107

Note:

• Since the tests uses abseil, with class based tests, I have removed the class which it subclasses to unittest as there is better compatibility of pytest with unittest. Removing all the class based test wouldn't have been wise as that would require finding compatible equivalents and replacing all instances, which should better be done in parts instead of in one go, if the team decided to get completely rid of abseil.

• Removed flags from abseil to use pytest addoption

[peytondmurray]

We also need the part of the workflow that automatically uploads to PyPI when a release is made. See https://github.com/tensorflow/tfx/blob/master/.github/workflows/wheels.yml for an example of how to do this.

[peytondmurray]

Usually I like to split builds and testing because they're separate concerns, but in this case building takes a LONG time if I remember correctly so I could be convinced here that this is a better approach. The only other thing is that absltest.TestCase is a child class of unittest.TestCase, so I'd be interested to know if replacing absltest.TestCase is strictly necessary; otherwise I think we don't need that change. But I don't think this should block merging.

[peytondmurray]

A few things remain:

• revert unit test/absltest change
• what do you think of putting pytest.ini content in pyproject.toml? It's a good place for project settings, and would cut down on the number of config files we have...

[peytondmurray]

Awesome, thank you!

[XinranTang]

Hi, could you resolve the conflicts in this branch?

[aktech]

Hi, could you resolve the conflicts in this branch?

Done, thanks!