Fix CUDA Runtime CTK test fixture on Windows ARM64

[isVoid]

What changed

Update the Windows CUDA Runtime CTK test fixture to model the current CUDA 13 layout explicitly:

• x64 creates bin/x64/cudart64_13.dll;
• Arm64 creates bin/arm64/cudart64_13.dll;
• Linux behavior is unchanged.

Why

#2393 made Windows CTK dynamic-library searches architecture-aware. Native Arm64 Python now correctly searches bin/arm64, but _create_cudart_in_ctk still created cudart64_12.dll in the legacy unqualified bin directory. As a result, test_try_via_ctk_root_regular_lib failed only under native Windows Arm64.

The fixture now targets the latest CUDA layout for both Windows architectures. Production discovery continues to retain its legacy unqualified bin fallback for older x64 toolkits.

Validation

• Native Windows Arm64 focused regression: 1 passed.
• tests/test_ctk_root_discovery.py: 37 passed.

[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.

[isVoid]

/ok to test 2b21582

[isVoid]

/ok to test 462c571

[isVoid]

Ah, closing in favor of duplicate PR: #2528

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2534/
https://nvidia.github.io/cuda-python/pr-preview/pr-2534/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2534/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2534/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.