Distributed solves

All optimizers now work with MPI. The mesh is currently partitioned into one domain per rank (1 rank is valid) and uses distributed MUMPS to solve the subproblems. If there are more than 1 subdomain, each output is saved in a split format (file ending with _rank_n) Next step would be to add iterative solvers, which is still WIP. Some memory leaks still occur and will be fixed soon !