Defer xfail condition evaluation with xfail_if_raises context manager

[elovelan]

Summary

This PR introduces xfail_if_raises, a context manager that approximates @pytest.mark.xfail(..., raises=...) behavior inside a test body rather than as a decorator.

Problem

@pytest.mark.xfail evaluates its condition argument during test collection (module import time). In test_index_mutation, this condition includes Git().config("core.symlinks") == "true", which:

1. Spawns an external git config process even when the test is not being run
2. Can fail or behave unexpectedly during collection if git configuration is unavailable
3. Relies on in-repo code during test discovery, which may not be appropriate in all contexts

Solution

xfail_if_raises moves the xfail logic into the test body. The side-effectful condition is only evaluated if the test actually raises the specified exception(s). This keeps test collection lightweight and avoids relying on in-repo code as part of module import/test discovery.

This change is applied to test_index_mutation in test/test_index.py.

AI disclosure

This PR description was prepared with assistance from an agent. All code was written by hand.

[elovelan]

I didn't change anything that should have affected this one failed test on Cygwin. I have a Windows VM now so I'll see if I can repro.

[EliahKagan]

The failed Cygwin test looks like a known flake with submodule ownership on Cygwin CI that I've been working on and hope to have a PR in for soon. To the best of my knowledge, that is independent of the changes here, though I emphasize that I have not reviewed this PR. I've triggered the test to rerun--my guess is that it will pass.

[Byron]

Thanks a lot, looks good to me!

Initially I was confused about strict as it's not actually used, but it's Ok and nothing one couldn't figure out and address next time someone gets confused by it.

[elovelan]

Initially I was confused about strict as it's not actually used, but it's Ok and nothing one couldn't figure out and address next time someone gets confused by it.

I debated a YAGNI approach, but I figured as a utility method, I should try to mirror the decorator as closely as I could.