scipy: drop the hypothesis pin now that we ship a cp314t wheel

[luhenry]

Removes the CIBW_TEST_REQUIRES override that pinned hypothesis<6.156.

That pin was a workaround for a gap that no longer exists. hypothesis became a Rust/maturin extension at 6.156, and upstream builds riscv64 abi3-only — cp310-abi3 plus a cp315-abi3.abi3t — so nothing covered cp314t (the stable free-threaded ABI starts at 3.15). On that interpreter pip fell back to the hypothesis sdist and the Rust build died computing riscv64-unknown-linux-gnu, a triple rustup does not have. That is what failed scipy's cp314t job. We now publish hypothesis 6.165.10 for cp312/cp313/cp314/cp314t (#451), so the pin is unnecessary.

The override restated upstream's [tool.cibuildwheel] test-requires verbatim apart from the pin, so it is deleted outright rather than unpinned in place — scipy's own pytest pytest-xdist threadpoolctl pooch hypothesis is inherited again, and there is one less place to drift from upstream.

hypothesis is added to PIP_ONLY_BINARY so resolution has to find a wheel. If our registry copy is ever missing or falls behind PyPI's latest, the job fails on "no matching distribution" instead of quietly rebuilding the sdist and failing later with a rustup error that names the wrong problem. It is in CIBW_ENVIRONMENT rather than CIBW_TEST_ENVIRONMENT because that variable is already there for numpy and scipy-openblas32, and hypothesis is not a build requirement, so covering both phases costs nothing.

Verified on the registry before pushing: hypothesis-6.165.10-cp314-cp314t-manylinux_2_34_riscv64.manylinux_2_39_riscv64.whl is present, alongside cp312/cp313/cp314; and scipy 1.18.1 itself is published for all four interpreters.

🤖 Generated with Claude Code