The old one unnecessarily used two separate queries for Postgres
& MySQL. And it had a bug in how it used MySQL's extended
"GROUP BY" functionality. While MySQL does let you return
non-aggregated columns, it uses an arbitrary record for these of
them, NOT the first record according to ORDER BY.

Also restructured the query a bit for optimization purposes.

behatnotneeded (Test in subsequent patch)

Change-Id: I3d49292025dcf7ac510a84d7e800476defec0525