Support using the static `libcudart_static` for the CUDA plugin.

[trivialfis]

The static library can serve as a stub to avoid linking a specific CUDA runtime version, which can help us distribute a more portable binary for the pyhwloc.

For CMake, one can link the CUDA::cudart_static CMake target, and it should figure out the rest, including the -ldl used by this stub library. I'm not sure how to do it for the autotools build.

[bgoglin]

Can you implement it (as an option?) in the CMake stuff? and we'll keep the issue open for autotools in case somebody wants it.

[trivialfis]

Yup, I can try to update the cmake build.