Make the "ORDER BY ... LIMIT n" optimizer take into account condition
selectivity data from EITS (not just from potential range accesses).