Test and CI updates

[kandersolar]

Description

This PR:

• Drops python 3.6 and adds python 3.9, 3.10 to CI tests
• Some tests currently use deprecated pvlib functions. This PR adds equivalent tests using non-deprecated pvlib functions. Each set of tests is marked to only run if the appropriate pvlib version is installed. The older tests can be deleted when we eventually drop support for pvlib<0.9.0 down the line.
• Resolves a few pandas deprecation warnings in the test functions. Most of the pandas warnings are not easily fixed and are not addressed here.

Because this PR doesn't add any new tests using multiple Arrays, it only partially addresses #116.

Checklist

• [ ] Closes #xxx
• [ ] Added new API functions to docs/api.rst
• [ ] Clearly documented all new API functions with PEP257 and numpydoc compliant docstrings
• Adds description and name entries in the appropriate "what's new" file
  in docs/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`).
• [ ] Non-API functions clearly documented with docstrings or comments as necessary
• [ ] Added tests to cover all new or modified code
• Pull request is nearly complete and ready for detailed review

[kandersolar]

Ok, ready for review.

A bit of context for the last commit: bfc9390 fixes an issue with scipy v1.8.0 (released Feb 5, 2022), which includes a change where scipy.stats.linregress raises an error if all x values are identical (scipy/scipy#14757). In most cases this wouldn't matter, except some of the tests have all identical y values and it seems that quadratic_r2 passes in (y, x) instead of the (x, y) that scipy.stats.linregress requires:

pvanalytics/pvanalytics/util/_fit.py

Lines 66 to 68 in 6fb4c87

	_, _, correlation, _, _ = scipy.stats.linregress(
	y, quadratic(x)
	)

However, switching to the correct parameter order doesn't get all the tests passing: quadratic(x), when y has zero variance, returns identical values and causes the scipy error again. I've added a special case for this to return r2=0 when y has no variability.

[cwhanse]

OK to merge when the whatsnew conflict is resolved.

[cwhanse]

Build fails after merging #127. But it appears the failures will be fixed with this PR.

[kandersolar]

Build fails after merging #127. But it appears the failures will be fixed with this PR.

Tests here are green, so I expect so.