TST: restore global state modified during tests

[yut23]

PR Summary

Restore numpy's error handling settings and yt's logging level during module teardown.

I found that test_fisheye in yt/visualization/volume_rendering/tests/test_vr_cameras.py was failing when I ran just the visualization tests, but not the full test suite.

PR Checklist

• New features are documented, with docstrings and narrative docs
• Adds a test for any bugs fixed. Adds tests for new features.

[neutrinoceros]

good catch, thank you !
here are a couple minor comments but I'd merge it as is

[neutrinoceros]

I'm not a fan of these all living on the same line. No harm done in keeping the existing one (though I'd also approve expanding it) but I'd rather have the new binding on a separate line.

[neutrinoceros]

I'd merge it as is

well... when tests pass, I guess 😅

[yut23]

Huh, I have no idea why only py3.9 is failing. I wasn't able to get test_fisheye to pass locally on 3.10, 3.11, or 3.12.

[neutrinoceros]

I think I saw this error being flaky a while back in a different context. If I recall correctly this was due to some uninitialised memory being involved in an array division, which may trigger this RuntimeWarning but only sometimes.

[yut23]

I tried running under valgrind, and the test passed with no memory errors. I'm pretty sure valgrind is just swallowing the FPE signals before numpy can handle them, though; I broke into a pdb session and the image array is all zeros with and without valgrind. I also checked against yt 3.6.1 and got the same error.

[yut23]

Alright, adding log_fields=[False] makes it produce a non-empty image. Should I make a separate PR for that?

[neutrinoceros]

Should I make a separate PR for that?

I don't think that's necessary.

@yt-fido test this please

[neutrinoceros]

@yt-fido test this please