use Popolo's sort_name for sorting candidates alphabetically

[mhl]

At the moment there are various places (see git grep 'split()\[-1\]') where we pick the last white-space separated part of the name to sort on. This doesn't work even in the UK, where some people have double-barrelled surnames that aren't hyphenated - in other countries, where sorting conventions vary, it's even less likely to be right.

Popolo has a sort_name field which we should use instead. To use this we would need to:

• Add a data migration that adds a SimplePopoloField for sort_name, and for each person populates sort_name with name.split()[-1] as a default.
• Add a post_init signal handler to suggest the last name as a default whenever a Person object is initialized, as in Pombola: https://github.com/mysociety/pombola/blob/4a66320295a74d0e2868bc431fac43686d3f3a59/pombola/core/models.py#L542-L548
• Change the instances where split()[-1] is used as a sort key to use sort_name instead
• Either add sort_name to the CSV columns that are output or fix this problem teach CSV_ROW_FIELDS and PersonExtra.as_list_of_dicts about custom fields #915 (the latter preferred)

[wfdd]

This is still not enough for sorting to work properly in other languages; the various built-in sort functions operate on bytes (i.e. Unicode code units), the order of which may or may not correspond to alphabetical order - see #937.