MPI CUDA-aware workaround

Add compile option when MPI-Cuda aware is not available.

Implementation:

• Add Kokkos::DualView to exchange ghost cell buffers via the host instead of directly between devices.
• MPI-Cuda aware will be turned off by default. If it is turned on by the user but not found by the code a warning will tell the user. Be aware that warning is not always reliable because it relies on a grep using ompi_info.