Add pretransform for full-sphere selection

[Carifio24]

I realized that I missed one thing in #2348 with regards to selection. Any points whose latitudes were wrapped back into [-π,π] won't be selected when making a ROI selection because the scatter viewer's apply_roi doesn't know about that transformation. This PR adds a pretransform that lets it know about what happens with longitudes.

[dhomeier]

Thanks for the fix and improving the tests!
I could not read out of the test setup if there is actually an example with, say RA > 2*pi input, tested. Otherwise only the small restructuring request.

[dhomeier]

Suggested change

	for angle_unit in ['radians', 'degrees']:
	for angle_unit in ['radians', 'degrees']:

I think testing structure would benefit from instead setting this and the other test up with

@pytest.mark.parametrize('angle_unit', ['radians', 'degrees'])

[Carifio24]

I agree, using parametrize would be better. Maybe should include the expected mask as a parameter as well?

[dhomeier]

As in parametrize('angle_unit,expected_mask' (...)? Had not even thought of that, but yes, that might make it even clearer.

[Carifio24]

@dhomeier I've updated the tests to use parametrize like we discussed, and have the full-sphere test also check the expected masks. I also made sure that some of the data points require longitude wrapping as you had mentioned, at least in radians mode.

Some of the export tests are now failing. I suspect these failures are unrelated because they're using the rectilinear projection, but my local tests are passing and I haven't had time yet to investigate further.

[Carifio24]

The same errors with the image export are popping up in the CI of other PRs (e.g. #2365), so I think it's safe to say that they're unrelated.

Edit: one of the errors was related, which has been fixed.

[Carifio24]

Just rebased to pick up the changes from #2370 for CI purposes

[dhomeier]

Looks all good now; I trust that the tests have the correct expected outcomes, thanks!

[dhomeier]

Was wondering why you switched this, as the conversion from degrees should come first, but now realised the pretransforms are applied in reverse polish notation...