Enable Windows CUDA CI

[zfergus]

Description

This PR adds a Windows leg to the CUDA workflow and fixes CI issues that the new leg exposed in cuda.yml, continuous.yml, and the nvcc warning flags.

• Per-OS CUDA install: On Linux, setup-pantheonsim (from CI: run the GPU unit tests on a simulated GPU #265) installs CUDA 12.8 with a simulated T4 and runs the GPU tests. On Windows, Jimver/cuda-toolkit@v0.2.35 installs CUDA 13.2, and the job only builds, since PantheonSim only supports Linux.
• Windows toolchain: Ninja + MSVC (msvc-dev-cmd), with CMAKE_CUDA_COMPILER passed explicitly because check_language(CUDA) doesn't reliably find nvcc on the runner.
• Matrix: Linux builds Debug and Release. Windows builds Debug only, because parallel Windows toolkit installs conflict.

Impact:

• Windows builds of the CUDA code now run in CI, which catches MSVC-host nvcc errors that the Linux build never sees.
• CI now builds against both CUDA 12 (CCCL 2) on Linux and CUDA 13 (CCCL 3) on Windows.
• No library or API changes.

CI fixes

• Build failures could pass the step: The Build step now runs under bash. Under the Windows default, pwsh, a failing cmake --build followed by ccache --show-stats reported success.
• Linux-only steps on Windows: The x86-64-v3 CPU check and the GPU Tests step are now limited to Linux.
• ccache on Windows: The Windows Debug configure drops MSVC's /Zi, which ccache can't cache, the same way Linux drops -g.
• Deprecated action: Removed seanmiddleditch/gha-setup-ninja@master from both workflows. The action is deprecated, and Ninja ships on all GitHub-hosted runners.

Build change

• With an MSVC host, nvcc's front end reports two warnings inside spdlog's bundled fmt: Kinetics and Static Friction pairwise for different materials #128 "loop is not reachable" and #2417 "constexpr constructor calls non-constexpr function". This happens despite the -isystem include, and -Werror all-warnings turns both into errors.
• ipc_toolkit_warnings.cmake now passes -diag-suppress=128,2417 for MSVC hosts only. The Linux build still treats both as errors, so our own code is still checked.
• -Werror all-warnings also turns MSVC's own host-pass warnings into errors, so #pragma GCC in lbvh.hpp failed on C4068. The MSVC /wd list is now shared and forwarded to nvcc's host pass with -Xcompiler, matching the C++ build.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Enhancement (non-breaking change which improves existing functionality)

How Has This Been Tested?

• CI only; Windows CUDA can't be built locally.
• Linux CUDA (Debug) passed on CUDA 13.2 before the merge with CI: run the GPU unit tests on a simulated GPU #265.
• Windows CUDA (Debug) previously failed on the fmt warnings, then on C4068. Both fixes are in f7e4c0779, and its run is pending.
• macOS configure re-run locally to check that the non-MSVC warning setup is unchanged.

Test Configuration:

• OS and Version: ubuntu-latest, windows-latest (GitHub-hosted)
• Compiler and Version: nvcc 12.8 + GCC (Linux), nvcc 13.2 + MSVC (Windows)

Checklist

• I have followed the project style guide
• My code follows the clang-format style guidelines of this project
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.92%. Comparing base (bdabc2d) to head (f7e4c07).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #262   +/-   ##
=======================================
  Coverage   96.92%   96.92%           
=======================================
  Files         194      194           
  Lines       17349    17349           
  Branches      943      943           
=======================================
  Hits        16816    16816           
  Misses        533      533           

Flag	Coverage Δ
unittests	96.92% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.