get_h5_value: use 'h5py_read_dataset' and avoid redefining it upper.

close #88

assigned to @payno

I would really prefer to keep val_ptr = fid[h5_path][()] rather than val_ptr = h5py_read_dataset(fid[h5_path]). With the former, we simply use the h5py/HDF5File API.

With the latter, we use yet another layer of libraries quite artificially, for little benefit. Even worse, we saw catastrophic failures of silx and HDF5 handling in the past release, so I'd rather stay with the simplest call as possible

On the other hand:

• This was the implementation of the

# TODO: look at silx.io.utils.h5py_read_dataset might replace it

• it removes local handling of the decoding (less code to maintain). and move to something more robust in my opinion (the h5py_read_dataset should handle more use case if the value change with time)
• h5py_read_dataset is already heavily used. So if it break having it at one more / one less place would not make much difference.

Regarding the bad time we had during last release I understand the reluctance, but from my point of view with it we should have something more robust / maintain with less code. (at the condition the upper code is properly handling 'converted' dataset. This I don't know).

Nevertheless I won't fight more for it. Else we can remove the TODO.