Refactor standalone Windows Nsight discovery

[rwgk]

Description

Closes #2609

Extract standalone Windows Nsight product discovery from the generic binary utility finder into cuda.pathfinder._binaries.windows_nsight.

• Move Nsight-specific registry lookup and architecture-specific executable layouts into the dedicated module.
• Expose lazy nsys and ncu candidate iterators while keeping candidate resolution, executable validation, absolute-path normalization, caching, and search orchestration in find_nvidia_binary_utility.py.
• Preserve launcher-first ncu.bat handling without eagerly detecting the native Windows machine architecture.
• Reorganize registry and Nsight-layout tests along the new module boundary while retaining public-finder and search-policy coverage.

This is a semantics-neutral internal refactor. It does not change the public API, candidate precedence, caching, error behavior, path normalization, or observable search order. The existing 1.6.1 release note already covers the user-visible Windows Nsight discovery behavior, so no documentation change is needed.

Testing

• TestVenv/bin/python -m pytest cuda_pathfinder/tests (1324 passed, 4 skipped)
• pre-commit run --all-files

Checklist

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

[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

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2614/
https://nvidia.github.io/cuda-python/pr-preview/pr-2614/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2614/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2614/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.