Possible wrapper bugs in __fenix_repair ranks

[rfvander]

In the following code. I think we need to use PMPI, not MPI, in the calls between pairs of asterisks. If not for set_error_handler, then certainly for free, because this will be something we will need to intercept in the future.

while (!repair_success) { 
    repair_success = 1;
    ret = MPIF_Comm_shrink(*__fenix_g_world, &world_without_failures);
    /* if (ret != MPI_SUCCESS) { debug_print("MPI_Comm_shrink. repair_ranks\n"); } */
    if (ret != MPI_SUCCESS) { 
      repair_success = 0;
      goto END_LOOP; 
    } 
 
    ret = **MPI_Comm_set_errhandler**(world_without_failures, MPI_ERRORS_RETURN);
    if (ret != MPI_SUCCESS) { 
      repair_success = 0;
      **MPI_Comm_free**(&world_without_failures);
      goto END_LOOP; 
    }

[keitat]

I will apply a fix to issue_01 and master copy.