feature: Support building CUDA backend kernels with clang++

[raffenet]

A user reported being unable to build MPICH using clang++ as the CUDA compiler used to build Yaksa pmodels/mpich#6954. Noting down the issues encountered trying to build like this:

1. NVCC_FLAGS is ignored when supplying an alternative NVCC in the configuration environment. Addressed in backend/cuda: Use NVCC_FLAGS with user-provided NVCC #250.
2. CUDA_SM code generation flags from configure are not compatible with clang++. Need to use the right flags based on the compiler used.
3. Building .cu files with clang/clang++ puts the compiler in C++ mode. CUDA kernels with the C _Bool type cannot be compiled since that type is not in C++. C++ has a bool type, but it is not guaranteed to be compatible, so we need some special handling to do the right thing.
4. clang++ warns that the -Xcompiler option used by cudalt.sh is unrecognized. We should omit it when using clang++.

[hzhou]

Does clang work or does it have to be clang++?

We should standardize it between MPL and yaksa, maybe as a Autoconf macro in a .m4 file. We can manually copy to yaksa for now.

[raffenet]

Does clang work or does it have to be clang++?

We should standardize it between MPL and yaksa, maybe as a Autoconf macro in a .m4 file. We can manually copy to yaksa for now.

I tried NVCC=clang and it produced the same error about _Bool not defined.