How often each kind of field was abstained on (qa[].question, …), summed over
the records, keys sorted — where a heal-loop session pays off most, and, saved
over time, which field drifted.
Records with abstentions whose catalog row predates the field breakdown, so
they are missing from abstained_by_field; ka reindex adds them.
Of those, records whose only holes are fields the parliament never provides (onlyKnownGaps).
Records with at least one abstained field.