Add support to using more than one MPI rank per ensemble member

[Rombur]

This PR enables us to use multiple processors per ensemble member. This only works if the all the ensemble members use the same number of processors. This ensures that the mesh is partitioned the same way for all the ensemble members which simplifies some operations. We could remove this condition in the future.

The changes in the code are pretty small if you hide the whitespace changes. Most of the bugs were related to some data not being distributed to all the processors.

[Rombur]

This requires #268. Otherwise we get too many warnings and the compilation stops

[stvdwtt]

@Rombur, got it, I just merged in #268.

[Rombur]

Retest this please

[Rombur]

Retest this please