allowing chunked data in HDF5 files

Prior to fix, "OSError: Can't prepare for writing data (Can't convert from size
to size_i)" would be raised when reading models with the mpiio driver if models
where too big [1]. Solution was to allow chunks when creating the datasets.

[1]: https://stackoverflow.com/questions/27984813/writing-reading-large-files-with-hdf5-and-mpi-using-1-process-from-python