Re-add CUDA+Clang build in CI

I could not figure out how to make any of the CUDA 12+ with Clang work. They mostly resulted in ptxas issue for the `triangulated_surface` benchmark. Disabling benchmark made things compile, but things still failed in testing for a different reason.

CUDA 12 requires clang 16+. Tried CUDA 12.5.1 + Clang 18-19 (coming with the distro), did not work. 