Use conda-forge for all conda CI tests

[kandersolar]

• [ ] Closes #xxxx
• I am familiar with the contributing guidelines
• [ ] I attest that all AI-generated material has been vetted for accuracy and is in compliance with the pvlib license no AI
• [ ] Tests added
• [ ] Updates entries in docs/sphinx/source/reference for API changes.
• [ ] Adds description and name entries in the appropriate "what's new" file in docs/sphinx/source/whatsnew for all changes. Includes link to the GitHub Issue with :issue:`num` or this Pull Request with :pull:`num`. Includes contributor name and/or GitHub username (link with :ghuser:`user`).
• [ ] New code is fully documented. Includes numpydoc compliant docstrings, examples, and comments where necessary.
• Pull request is nearly complete and ready for detailed review.
• Maintainer: Appropriate GitHub Labels (including remote-data) and Milestone are assigned to the Pull Request and linked Issue.

The windows python 3.14 conda job has been failing for a while with a mysterious error:

Error: The process 'C:\Users\runneradmin\micromamba-bin\micromamba.exe' failed with exit code 3221225477

https://github.com/pvlib/pvlib-python/actions/runs/36232852407/job/108379157046

I couldn't find anything in the setup-micromamba and mamba repositories. I tried lots of debugging in my fork's CI, for example getting rid of all the dependencies and just installing python, changing the windows version back to 2022, changing the mamba version, changing the setup-micromamba version... nothing worked, except for disabling the main anaconda channels and using only conda-forge. So, that's what this PR does. Does that reduce the value of our tests in any way?

[kandersolar]

As noted in #2876 (comment), it seems that the issue stems from mamba supporting only conda-forge (not the defaults channel). So let's proceed with using only conda-forge in the tests.

CI is green, so merging.