Add support for building fatbin from clang ptx

[rbberger]

This enables building the embedded fatbinary using ptx generated by Clang. It is currently restricted to a single .cu file in $REALM_CUDA_SOURCES and a single $CMAKE_CUDA_ARCHITECTURES. The single .cu restriction is already the case also for the NVCC path anyway.

For more SMs per translation unit we would have to either ask CMake to fix their CUDA_PTX_COMPILATION support, or manually do it via loops and custom targets/commands. For the FleCSI use case both restrictions are currently sufficient.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 27.74%. Comparing base (23ab773) to head (5166709).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #368      +/-   ##
==========================================
- Coverage   27.74%   27.74%   -0.01%     
==========================================
  Files         192      192              
  Lines       39502    39502              
  Branches    14382    14382              
==========================================
- Hits        10960    10959       -1     
  Misses      28151    28151              
- Partials      391      392       +1     

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