Avoiding MPI_Comm_compare altogether in the common case

[marcgamell]

What I suggest is to optimize for the common case. In the common case we don’t need to call MPI_Comm_compare at all. Therefore, we can convert the code in PMPI functions to something like:
If(use_hack) {
If(comm___ == comm) Use resilient_comm;
Else Use comm;
} else {
MPI_Comm_compare(dup_of_input_comm, comm);
If(CONGRUENT) Use resilient_comm;
Else Use comm;
}

We can determine in Fenix_Init if use_hack is appropriate or not. For OpenMPI (and probably MPICH, once ULFM is implemented in it), use_hack will always be true. This would solve the AMR case in which many different comms are used.

[marcgamell]

George confirmed that direct comparison should work for all libraries:
For MPI_IDENT a direct comparison with == should work for all libraries.