One row per recipient_uid, choosing the most complete representative of
each cluster (longest name, then most address fields populated) so the match
runs against the best available spelling rather than an arbitrary one.
Arguments
- grants
Output of
pf_dedup().