prepared implementation of parallel dumpers

Currently, only the H5Dumper is MPI-compatible, provided h5py was installed with
MPI support (https://docs.h5py.org/en/latest/build.html#custom-installation).

The UVWDumper needs a lot of work, because VTKRectilinearGrid supposes overlap
of nodal values at proc boundaries. Communication of overlap is necessary :/

The NetCDFDumper needs testing against a parallel netcdf library.