[Feature Request] Generic Parallel HDF5 Improvements

[jhendersonHDF]

The following were noted as general high-level ideas for improvements to Parallel HDF5:

• Adding support for independent metadata modifications
• Fixing space allocation for chunked datasets in all cases so that all the chunks don't need to be allocated in advance when accessing a file with parallel HDF5. Note that this is currently supported in parallel HDF5 only if filters are applied to the chunked dataset.
• Adding support for being able to convert between an MPI derived datatype and an HDF5 selection. Note that there should be less or no need for this as the library moves towards using vector I/O to eliminate passing down opaque MPI derived types to VFDs.