Add Windows/MacOS to test-sundials CI

[c-randall]

This PR adds a build-and-test routine to test-sundials.yml for both Windows and MacOS to improve catching breaking changes on various operating systems, as suggested in PR #196.

Notes:

• I'm not aware of a convenient way to build SUNDIALS using a variety of precisions and index sizes on Windows or MacOS, so I used builds from conda-forge (double precision, 32-bit index).
• You may want to consider adding macos-13 to the matrix.os section. macos-latest uses arm64 architectures and macos-13 uses x86_64, so testing both would likely be more robust.
• I only added Python 3.9 and 3.13 to the matrix.python-version section since those are the oldest and newest versions that are currently supported. Feel free to add more.

Comments:

I noticed that the tests inside packages/scikits-odes-sundials/src/scikits_odes_sundials only test the _get_num_args function, so the tests here are not particularly robust. Now that scikits-odes-sundials can be distributed as its own package, it may be worth moving over some of the more rigorous tests that are specific to odes-sundials from packages/scikits-odes/src/scikits/odes/tests.

[aragilar]

Can you change this to use tox, so any changes that are needed to be made can be centralised in tox (and hence avoid coupling to the specific CI system)?

[c-randall]

I changed the new CI job to use tox, but I had an issue with the Windows and Python 3.9 combination (see failed actions). The virtual environment that tox created had an issue with pip and SSL. I tried tox --recreate and I tried getting tox from conda instead of PyPI, but neither "fix" worked. For now, I excluded this combination from the tests.

This seems to be an issue with how the runner sets up its virtual environments with tox. The previous method was working on Windows with Python 3.9, and I verified that the new tox method works on my local machine (Windows) in a virtual environment with Python 3.9, but I cannot figure out how to resolve the issue on the runner.

After this PR is closed, can you also do a v3.1.1 release that includes the fix from PR #196? Even though the PR was closed, there was no release to push the patch to PyPI.