AttributeSubsettingIterator currently executes prior to QueryFilterIterator (if a subset of fields is provided).
It should be reworked to execute after QueryFilterIterator, so any filtering happens first, and attributes are only filtered from the results that pass all filters. This should result in a performance boost for this use case. Specifically, AttributeSubsettingIterator will need to be modified to operate on encoded "whole rows" rather than individual key-value pairs.