Aggregation of fields referencing external variables

[sadielbartholomew]

Implement agreed behaviour for the aggregation of field constructs having at least one reference to an external netCDF variable to define cell measure construct(s) (the only possibility allowed by the current status of the CF Conventions), namely to aggregate such fields otherwise aggregatable but to delete the external cell measure on the aggregated output.

This is sensible behaviour that will be sufficient for the next release, 3.8.0, but we decided it would be good to review the meaning of external cell measures towards a consistent and well-defined treatment in future (see Issue to be raised and linked to this PR).

Note this closes #150 (see in particular the comment here) since it enables the fields there to be aggregated to one.

[davidhassell]

This looks very good to me. We should hold off merging just to conclude the discussion on https://github.com/NCAS-CMS/cf-python/pull/172/files#r544996379.

[sadielbartholomew]

Hi @davidhassell, I have just finalised the PR to take the revised approach as you suggested in the comment above. A test was amended slightly to account for the new behaviour (the field split into three parts with one having had the external measure deleted will aggregate to a single field rather than two that would in turn aggregate to one if run through cf.aggregate again, as desired.) So if you are happy this is ready to merge.

Sorry for the delay, since the Telco I have been working on the append mode PR (locally) in parallel...

[sadielbartholomew]

(Since my comment above I saw I had one changelog conflict to resolve, so have just done that.)

[davidhassell]

Looks good, thanks - merge away!