Add calls to smpi_bench_{end,begin} in PMPI_Wtime.

Add calls to smpi_bench_{end,begin} in PMPI_Wtime.

Without this change, doing something like the following always
gives a duration of zero:

    ...
    start = MPI_Wtime();
    // big computation...
    end = MPI_Wtime();
    duration = end - start;
    ...

Make sure that no calls to smpi_bench are made when process is not
between init and finalize, nor during sample blocks.