build(release): prepare Windows CUDA wheel metadata

[IvanaGyro]

Stack

#1109 → #1110 → #1116 → #1117 → #1118 → #1119 → #1113

Depends on #1118 and is the final implementation layer before the workflow-only #1113.

Summary

• teach the CUDA release-manifest rewrite to emit Windows PyPI runtime dependencies and disable unavailable cuQuantum packages;
• preserve the existing Linux RUNPATH/auditwheel rewrite;
• add platform-specific release-manifest tests and the release-only tomlkit dependency;
• enable IPO for release wheels while leaving it disabled in development tasks.

Linux and macOS

The Linux rewrite remains the default and is covered by the same focused tests. macOS configuration is untouched.

Validation

• pytests/prepare_cuda_release_test.py: 3 passed;
• locked wheel-cpu environment installs with the release-test dependency;
• repository-pinned pre-commit hooks pass.

Part of #1114.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4ade236e2c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep IPO off for editable development builds

When contributors run pixi run install-python or pixi run -e cuda install-python-cuda, those tasks configure separate *-editable build directories without an IPO override (lines 94 and 125). Since CMakeLists.txt:304-308 defaults CMAKE_INTERPROCEDURAL_OPTIMIZATION to true when undefined, removing -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=OFF here and from the CUDA activation makes these development installs use /GL/LTCG as well, substantially slowing iterative builds despite IPO being intended only for release wheels. Keep the override in the development tasks or environments and enable IPO specifically in the wheel build.

Useful? React with 👍 / 👎.