Enable parallel netCDF dataset reads

By introducing two new netCDF backends:

- For netCDF-4: `h5netcdf-pyfive` = `h5netcdf` with the `pyfive` backend
- For netCDF-3: `netcdf_file` = `scipy.io.netcdf_file`

we can enable parallel reading of netCDF datasets.

`h5netcdf` with the `pyfive` backend should become the new default read engine (it's currently the non-threadsafe `h5netcdf` with the `h5py` backend). 

PR to follow.