The aggregate names don't need to be sorted, but it helps to do
this for the tests, since the implementation of sets may not be
stable between machines and lead to sporadic test failures.
I did also toy with sorting granular area names so they have a
similar ordering, but this would take them out-of-order with IDs
and really the important thing is that the ordering is stable.
I did consider whether to store this explicitly in the SQLite DB,
but this is less effort for now and we can always switch to that
more robust approach in future if we need to.
This applies some heuristics to try and keep the overall list of
areas short when many are selected in the same wider area.
Currently we only have relationship information between upper and
lower tier local authorities, so we can't / won't aggregate up to
Greater London (it's own special thing) or whole countries.
