Use shr_mpi consistently throughout CTSM rather than native MPI

Overall we use shr_mpi_mod.F90 from cesm_share in CTSM. There are some places that we use bare MPI and the code would be cleaner if we used shr_mpi_mod.F90 exclusively rather than sometimes not. We'd like to make this consistent which would make the code easier to understand and remove some unneeded complexity from having multiple ways of doing the same thing.