initial commit enabling GPU direct communication for Cray-MPICH

[JPRichings]

This needs testing but here is an initial suggestion for checking support for GPU enabled MPI in CRAY_MPICH this should enable direct inter GPU comms for a wider range of systems in the HPC space that don't always support OpenMPI for there scale out network.

Plan to test on MI210 nodes on ARCHER2 when they are available.

[JPRichings]

Should have updated this before:

Hardware: 1 node 4 gpus Mi210 ARCHER2

before (no MPI direct comms):

QFT run time: 9.49068s
Total run time: 10.2344s

after (with MPI direct comms):

QFT run time: 2.29135s
Total run time: 2.92993s

I think it might be worth switching on GPU direct comms... ;)

[JPRichings]

Test all failing due to:

[ 80%] Building CXX object _deps/catch2-build/src/CMakeFiles/Catch2.dir/catch2/matchers/catch_matchers_predicate.cpp.o
/Users/runner/work/QuEST/QuEST/quest/src/comm/comm_config.cpp:75:5: error: use of undeclared identifier 'MPI_Get_library_version'
   75 |     MPI_Get_library_version(version_string, resultlen);
      |     ^
1 error generated.

• I will look to add macro guard around this mpi call to fix non-mpi compilation.
• need to set some sensible default buffer sizes
• Need to remove the debugging messages for the code

[JPRichings]

need to return NONE for comm_whichMpi and false for comm_set_isMpiGpuAware

[TysonRayJones]

Maybe this should be isMpiGpuAware to be consistent with isGpuSharingEnabled, comm_set_isMpiGpuAware, etc