Improve data exchange in fastscape interfaces.

[bangerth]

As also found in #6723, the current fastscape interface has the following code:

      else
        // For ranks other than the root:
        {
          for (unsigned int i=0; i<local_aspect_values.size(); ++i)
            MPI_Ssend(&local_aspect_values[i][0], local_aspect_values[i].size(), MPI_DOUBLE,
                      /* destination is root= */ 0,
                      /* tag= */ 42,
                      this->get_mpi_communicator());
            ...

@Djneu says he doesn't remember why he used an Ssend here instead of the usual Send or Isend. That said, the corresponding receive is here, in fill_fastscape_arrays():

          // First, find out the size of the array a process wants to send.
          MPI_Status status;
          MPI_Probe(p, 42, this->get_mpi_communicator(), &status);
          int incoming_size = 0;
          MPI_Get_count(&status, MPI_DOUBLE, &incoming_size);

          // Resize the array so it fits whatever the process sends.
          for (unsigned int i=0; i<local_aspect_values.size(); ++i)
            {
              local_aspect_values[i].resize(incoming_size);
            }

          for (unsigned int i=0; i<local_aspect_values.size(); ++i)
            MPI_Recv(&local_aspect_values[i][0], incoming_size, MPI_DOUBLE, p, 42, this->get_mpi_communicator(), &status);

This is really just a gather operation: process 0 needs to obtain information from all other processes. We could save a whole bunch of code by just calling Utilities::MPI::gather(). I'll implement that once we have #6723 merged.