Prevent read error for datasets including badly-encoded UGRID meshes

[sadielbartholomew]

Closes #315. For the need for a quite specific dataset to test against and lack of bandwidth before we are due to do the new release set, I am not adding a test to capture the bug, which I appreciate is not ideal but there is a test in the new cf-plot suite which exposed this, which I think is good enough*.

*This involves updating the code snippet for the appropriate cf-plot example, where the field indexing seems to have changed so that I've matched the relevant fields like so to update it to work with this branch, whereas before it hit the error covered in the associated issue on the initial read:

>>> f = cf.read("/home/slb93/git-repos/cf-plot/cfplot/test/cfplot_data/lfric_initial.nc")
There was a problem parsing the UGRID mesh topology variable. Ignoring the UGRID mesh for 'u1'.
There was a problem parsing the UGRID mesh topology variable. Ignoring the UGRID mesh for 'u2'.
>>> pot = f.select_by_identity("air_potential_temperature")[0]
>>> lats = f.select_by_identity("latitude")[0]
>>> lons = f.select_by_identity("longitude")[0]
>>> faces = f.select_by_identity("cf_role=face_edge_connectivity")[1]  # try second of two
>>> pot_crop = pot[4,:]  # reduce
>>> cfp.con(f=pot_crop, face_lons=lons, face_lats=lats, face_connectivity=faces, lines=False)