cuda.bindings: build 12.9 and 13.x selectively from main

[kkraus14]

Summary

• import the maintained CUDA 12.9 bindings tree from 238955935bd903ac72817c0dfdfe4f6a54ee6bb1 as cuda_bindings_12/
• apply the repository's current Apache-2.0 license text and SPDX policy to the imported package
• extend the selective-CI workplan from ci: activate dependency-aware package builds and tests #2467 with cu12 and cu13 variants while preserving its aggregate module gates
• build, test, document, and release the CUDA 12.9 and CUDA 13 bindings/metapackage pairs from main
• retain the 12.9.x branch for exceptional legacy-compatibility backports, while removing it from routine build and test artifact sourcing

Selective CI behavior

Change	Bindings built/tested	Matching cuda-python pair	Other bindings line
cuda_bindings_12/** source	CUDA 12	CUDA 12	reused from the #2467 baseline
cuda_bindings/** source	CUDA 13	CUDA 13	reused from the #2467 baseline
shared bindings consumers or CI infrastructure	CUDA 12 and 13	CUDA 12 and 13 as needed	n/a

The existing modules.bindings.needs_* rollups remain available to #2467 consumers. Per-major decisions live under modules.{bindings,core,python}.variants.{cu12,cu13}, and wheel/sdist matrices are pruned with the corresponding job-level major gates. Core's two build halves are rebuilt together when either bindings line changes so the merged core wheel always comes from one source revision.

Stable v12.9.* tags select only the CUDA 12 bindings/metapackage pair, while v13.* tags select only the CUDA 13 pair. Exact configured CUDA 12.9 and CUDA 13 test rows consume current-run artifacts; older compatibility rows continue to exercise published wheels.

Validation

• repository pre-commit suite (SPDX, generated seals, Ruff, YAML/TOML, RST, large-file policy, and actionlint)
• pixi run python -m unittest ci/tools/tests/test_compute_ci_plan.py (6 tests)
• release-note routing tests (33 tests)
• CUDA 12 Pixi source build/import and Cython extension build on Linux with Python 3.14
• CUDA 12 and CUDA 13 cuda-python wheel metadata checks for line-matched dependencies
• current CUDA 12/current CUDA 13/older-minor environment-routing smoke checks

The full GPU wheel matrix is left to CI.

Depends on #2467.

Closes #1199.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.