Flexible MPI partitioning

So far the model has to be split amongst the MPI processes into bunches of equal size. This means that the number of grid cells (model size) must be an integer multiple of the processes in each direction. This poses a limitation on the split of the model and the number of MPI proceses that can be used (or the model needs to be changed in size). This could be avoided by help of packages like MPI cartesian grids.

Implement a variable MPI split of the model.