I noticed that filter_get_active_in_context in filterlib was using get_recordset_sql without a row limit, just because it was not very easy to get a unique value in the first column.

However, it is known than get_recordset_sql is almost always slower than get_records, because (on postgres at least), even for small results, it does two network round trips to the DB server, rather than just one.

Having re-witten the DB query in this function, it gives about a 20% speed-up, which seems worth having, so I am putting this up for review.