Fix a type in a MPI call in the mesh interpolation test. This makes it work also with more than one processor. Moreover, use local instead of global indices and simplify the mesh parallel handling.