Make readdict more consistent

With the solution !7 (merged) to #24 (closed), we now have a different definitions for readdict for HDF5 and JLD2, the latter with some copied-and-modief code. The generic readdict in InPartS is never used. In the long run, we should make this more consistent (for example, make the generic version more efficient), delete the generic version, or use some future functionality of JLD2 to get recursive Dict reading to work.