Add cuda-11.3+clang9 build workflow
To be able to detect unused captures in GPU code lambdas (as gcc does not support this diagnostic)
Remove unused opts lambda capture in `ProcessGroupMPI.cpp` and `Distributions.cu`
Fix sign-compare in nvfuser benchmark and ignore signed unsigned comparison in nvfuser tests
Fixes https://github.com/pytorch/pytorch/issues/75475 by aliasing CMAKE_CUDA_HOST_COMPILER to C_COMPILER when clang is used
Pull Request resolved: https://github.com/pytorch/pytorch/pull/75293
Approved by: https://github.com/atalman, https://github.com/seemethere
