Make xESMF optional

[aulemahal]

Pull Request Checklist:

• [x] This PR addresses an already opened issue (for bug fixes / features)
  • This PR fixes hydrologie/xhydro#81
• [ ] (If applicable) Documentation has been added / updated (for bug fixes / features).
• [x] (If applicable) Tests have been updated.
• [x] This PR does not seem to break the templates.
• [x] CHANGES.rst has been updated (with summary of main changes).
  • [x] Link to issue (:issue:number) and pull request (:pull:number) has been added.

What kind of change does this PR introduce?

• Make xesmf optional:
  • Remove it from the dependencies in "pyproject.toml". It is now an "extra" dependency. I did not remove it from the environment.yml files as the issue doesn't occur with conda.
  • Add a few "try: import" throughout the package and raise on call time when xesmf is not available.
  • Skip the tests that require xesmf, if not present.
  • Edit the tox config to remove any usage of conda, since it is now not needed.

Does this PR introduce a breaking change?

No.

Other information:

@Zeitsperre I may have been heavy handed in editing the workflow and tox configuration. My idea was that the "test-pypi" didn't need xesmf/esmf anymore and so I removed any mention of it and of mamba/conda. So it can really be a "test-pypi" ci.

[aulemahal]

Yes ça marche!

Issue with Python 3.9 is because of the latest conda package of intake-esm. They dropped 3.9 but the conda recipe didn't know about that.

[aulemahal]

CI fails because the misconfigured conda package still exists in mamba's cache and in the conda-forge repo. But it did pass when I pinned intake-esm.

Coverage for Pypi runs are decreased, which is normal since we don't test any function using xesmf. But the conda converage is stable. @Zeitsperre , what should we do ? Accept the decrease ? Stop reporting coverage for those tests ?

But otherwise, this PR is good to go.

[Zeitsperre]

@Zeitsperre , what should we do ? Accept the decrease ? Stop reporting coverage for those tests ?

I think accepting the one-time decrease is fine. So long as the overall coverage doesn't go down (test-pypi + test-conda), that's what matters.