[nychiang]

Description

Fix CUDA profiling builds with recent CUDA toolkits that no longer provide the legacy CUDA::nvToolsExt library target.

With CUDA 12.9, enabling RESOLVE_USE_PROFILING caused CMake generation to fail because libnvToolsExt is no longer shipped. NVTX3 is its supported header-only replacement.

No additional dependencies are required.

Proposed changes

• Prefer the CMake CUDA::nvtx3 target when available.
• Include <nvtx3/nvToolsExt.h> for NVTX3 builds.
• Export RESOLVE_USE_NVTX3 to downstream consumers.
• Retain CUDA::nvToolsExt as a fallback for older CUDA/CMake environments.

The change was built using CUDA 12.9.1 for SM90 and tested on an NVIDIA H100.

Checklist

• All tests pass (make test and make test_install per testing instructions). Code tested on
  • CPU backend
  • CUDA backend
  • HIP backend
• I have manually run the non-experimental examples and verified that residuals are close to machine precision. (In your build directory run: ./examples/<your_example>.exe -h to get instructions how to run examples). Code tested on:
  • CPU backend
  • CUDA backend
  • HIP backend
• Code compiles cleanly with flags -Wall -Wpedantic -Wconversion -Wextra.
• The new code follows Re::Solve style guidelines.
• There are unit tests for the new code.
• The new code is documented.
• The feature branch is rebased with respect to the target branch.
• I have updated CHANGELOG.md to reflect the changes in this PR. This is a minor CUDA compatibility fix, so no changelog entry was added.

Further comments

The following smoke tests passed:

• CPU: version, matrix, logger, and parameter tests
• CUDA: GMRES, sparse matrix, vector, and device-memory tests

The installed CMake package was also checked and correctly exports CUDA::nvtx3 and RESOLVE_USE_NVTX3.

[shakedregev]

Looks good and tested.