test(cuda.bindings): restore thread_unsafe marker for cuFile stats level

[rwgk]

Description

Restore @pytest.mark.thread_unsafe(reason="cuFile stats level is process-global") on test_set_stats_level.

#2782 added the marker because cuFile's stats level is process-global, but it was unintentionally dropped while reconciling the 13.4.x release branch into main in #2789. Without it, this test can race with other cuFile stats tests under parallel execution.

This restores the exact marker from #2782; no production code changes.

To keep CI viable while #2809 is awaiting merge, this branch also cherry-picks its exact cuTENSOR 2.8 test fix. Once #2809 lands, updating this branch from main will absorb those changes.

Testing

• Targeted pre-commit checks for the marker change: passed.
• test(pathfinder): handle cuTENSORMg removal in cuTENSOR 2.8 #2809's original head is fully green.
• pytest not run locally, per request; full CI will exercise the combined branch.

Checklist

• New or existing tests cover these changes. This change restores test scheduling metadata and includes test(pathfinder): handle cuTENSORMg removal in cuTENSOR 2.8 #2809's focused coverage.
• The documentation is up to date with these changes. No documentation changes are needed.

[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.

[rwgk]

/ok to test f77b79f

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2812/
https://nvidia.github.io/cuda-python/pr-preview/pr-2812/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2812/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2812/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

[rwgk]

/ok to test aac6277