MPI performance and n_walkers

Created by: SylvainMarsat

It seems (reported by Henri) that MPI performance depends significantly on whether the number of processes divides the number of walkers (*ntemps presumably). Clarify why. One option would be to use the schwimmbad MPI pool.