CUDA: Wrap `-pthread` flags into `-Xcompiler -pthread` for nvcc invocations
Using the finder module for [Threads](https://cmake.org/cmake/help/v3.11/module/FindThreads.html) with `CMAKE_THREAD_PREFER_PTHREAD=ON` causes compilation of cuda objects to fail with the message:
`nvcc fatal : Unknown option 'pthread'`
This is because nvcc does not understand the option `pthread`. This should be able to be made to work by wrapping it in an `-Xcompiler` option.
As a user-side workaround this is a bit unsatisfactory, because dealing with `-pthread` flags is platform dependent, whereas the whole point of the the FindThreads module is to abstract away the platform dependency.
What would be great is if the native CUDA support in CMake could handle interactions with `-pthread` natively.
Related to #17929.
Maybe @robertmaynard is involved, maybe the maintainer of the FindThreads module
issue