fix modelchain poa tests

[wholmgren]

• Closes #xxxx
• I am familiar with the contributing guidelines
• Tests added
• Updates entries to docs/sphinx/source/api.rst 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 and Milestone are assigned to the Pull Request and linked Issue.

As discussed here, test_prepare_inputs_from_poa didn't work when I ran the tests locally (as I expect from looking at the code) but the CI never failed (for reasons I still don't understand). This PR fixes that.

This PR also changes the modelchain.WEATHER_KEYS, POA_DATA_KEYS, TEMPERATURE_KEYS, and DATA_KEYS to tuples. The problem with using a set here is that it's unordered, so the _assign_weather and _assign_total_irrad create dataframes with column order that depends on the platform.

[wholmgren]

While we're at it, I wonder about the name POA_DATA_KEYS. Would POA_KEYS be more consistent with WEATHER_KEYS and TEMPERATURE_KEYS?

[cwhanse]

Were you able to figure out why the test passed the CI with that bad logic?

[wholmgren]

I was not able to figure that out, unfortunately. The fixtures are all function scoped either explicitly or by default, so I don't think they're the cause of the problem.