bump GraalPy versions tested in CI, test graalpy on windows

[davidhewitt]

Having some pain over in setuptools-rust in https://github.com/PyO3/setuptools-rust/actions/runs/34587139565/job/103223856532?pr=615 - hopefully bumping this fixes.

[codspeed]

Merging this PR will not alter performance

✅ 141 untouched benchmarks
⏩ 6 skipped benchmarks¹

---

_{Comparing davidhewitt:graalpy-win (eaa1e36) with main (9b41992)}

Footnotes

1. 6 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[davidhewitt]

@msimacek do you have any experience with pytest for GraalPy on windows? Looks like it's just exciting immediately with code 1 according to the CI log, I need to try to dig, not sure if you've seen similar before.

[msimacek]

pytest worked fine for me locally for this PR, but I also tried it on some other packages and ran into an issue that had the same symptom as yours. It could be worked around by turning off pytest's assertion rewriting (it's really weird that assertion rewriting would only be broken on windows, I'll need to look deeper into that). You can try if you get further with --assert=plain argument to pytest.

[davidhewitt]

It looks like hypothesis failing to build correctly (worked around in #6411 and fixed in this PR) was the root cause. Thanks for checking!

[ngoldbaum]

This also changes the library name when PYO3_USE_RAW_DYLIB=0. When my AI model tested, it broke graalpy because the library name is e.g. python312.lib but the build script requests python-native.lib anyway. I think this is correct only for PYO3_USE_RAW_DYLIB=1.

[davidhewitt]

xref #6410 (comment) - I guess it's the other side of the same hack, where a clause would be needed in that hack to also account for GraalPy.

I guess that means that only one of these PRs will merge successfully because #6410 introduces proper ffi-check for both modes of PYO3_USE_RAW_DYLIB.

I'll have a go at solving the full refactor properly, perhaps after merging #6410 but before merging here? Merging #6410 will unblock the rest of the expanded ffi-check testing in #6389