h5py: reading issue for groupBased file

[ax3l]

With h5py, both as backend in the current dev branch as well as using it through the last stable release, I have problems opening the following file through the slider:
openpmd_gzip.h5.gz

The openPMD-validator check does not raise errors on the file.

Since the problem does not appear with the dev branch using the openPMD-api backend, I think there might be a logic error in the h5py backend we need to address.

Errors in this notebook:
openPMD-visualization.ipynb.gz

cc @RemiLehe

Debugging Progress

GUI: The failing find_dataset call tries to interpret scalar records as vector records, e.g.:

find_dataset: <HDF5 file "openpmd.h5" (mode r)> - Bx/x
find_dataset: <HDF5 file "openpmd.h5" (mode r)> - EypBx/x

While things like Bx, By, ... will become vector records eventually - others like EypBx stay scalar records.

API call: Running

ts.get_field("EypBx", coord=None, iteration=9, plot=True)

gets the find_dataset right

find_dataset: <HDF5 file "openpmd.h5" (mode r)> - EypBx

but reads the zero (0) values from iteration=0 even if one requests iteration=9.