right now, we only drop the namespace entries for each child schema -- we need to drop the descriptors as well.