Bump Cython to >=3.3,<3.4

[juenglin]

Description

Bumps the Cython build/test pin from >=3.2.5,<3.3 to >=3.3,<3.4 across cuda_bindings, cuda_core, cuda_pathfinder, and docs environment files.

No motivating issue -- routine dependency upgrade.

Motivation

• Python 3.15 support path: Cython 3.3 adapts to 3.15 and its Limited API; our test dependency groups already carry python_version < '3.15' guards to be removed once 3.15 is officially supported.
• Free-threading improvements: critical sections on auto-generated functions, prange/parallel() without releasing the GIL, pymutex.locked().
• Runtime performance improvements in generated code: faster memoryview indexing/slicing, vectorcall-based extension type instantiation.

Required code changes

Cython 3.3 added a new critical_section misuse warning (cython/cython#6766), promoted to a build error by our warning_errors=True builds. It flagged three real patterns in cuda_core:

• cuda/core/_memory/_buffer.pyx, cuda/core/_memoryview.pyx: narrowed @cython.critical_section coverage to only the cdef-attribute check-and-set; Python-level construction/reads now happen outside the lock, since the lock never actually protected them (see cython/cython#7324 for upstream discussion of this exact limitation).
• cuda/core/_device_resources.pyx: WorkqueueResourceOptions now uses @dataclass(init=False) with a hand-written __init__, because Cython 3.3 unconditionally wraps dataclass-generated __init__ in a critical section and then warns about its own generated call to __post_init__.

Two cuda_bindings tests (test_eglFrame, test_cudart_eglFrame) were updated to pass CUarray/cudaArray_t objects instead of raw ints, because Cython 3.3 now enforces declared container item types (list[CUarray]) at runtime. The setter's type annotation comes from generated code; the code generator may want to revisit it separately.

Build times

Clean full-rebuild times (cache disabled, median of 3 runs) are essentially unchanged between Cython 3.2.9 and 3.3.0 for both cuda_bindings and cuda_core on this hardware.

Test parity

Identical results before/after on a local GPU (CUDA 13.2):

• cuda_bindings: 552 passed / 26 skipped (+ 9 passed in tests/cython)
• cuda_core: 3977 passed / 237 skipped / 4 xfailed (+ 1 passed in tests/cython)

Known follow-up (out of scope here)

Relocking pixi.lock for cuda_bindings/cuda_core to reflect the new Cython pin incidentally triggers a full re-solve of the unconstrained default environment, which also bumps its CUDA toolchain from 12.9 to 13.3 (pre-existing lock drift relative to ci/versions.yml, unrelated to Cython). That drift and its long-term fix are tracked separately in #2770; pixi.lock files are intentionally left untouched in this PR.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[juenglin]

/ok to test 6f24210

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2771/
https://nvidia.github.io/cuda-python/pr-preview/pr-2771/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2771/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2771/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.