Enable parallel netCDF dataset reads #912

By introducing two new netCDF backends:

• For netCDF-4: h5netcdf with the pyfive backend
• For netCDF-3: 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).

With pyfive in the mix, we can also improve the performance of active storage reductions, by passing extant pyfive Dataset objects to activestorage.Active, as opposed to a dataset URL and variable name.

PR to follow (which will require NCAS-CMS/cfdm#375 to be resolved first).