CMake: support CUDA-aware MPI
Currently, building with cmake does not support CUDA-aware MPI, which is a critical feature on many GPU systems.
With autotools, the behavior is as follows:
- if
--enable-cuda-mpi
is set, enable CUDA-aware MPI - if not, try an automatic detection for OpenMPI which is known to set a certain macro if it supports CUDA-aware MPI calls (see https://gitlab.com/octopus-code/octopus/-/blob/main/m4/cuda.m4?ref_type=heads#L84)