Make sure we add the new triggers after the "rename table, create
replacement one, dump data" dance we do on ontology changes. This will
avoid leaks as the last step there would increase refcount for all
involved resources.

There is only one situation where this does not work: Ontology changes
that remove properties with rdfs:Resource range will result in those
resources being leaked. This breaks differently but in similar ways for
single-valued and multivalued properties, in the first we fail to delete
elements from the old table, in the second we fail to delete the table
altogether.

This barely ever happens, so a FIXME has been added, we should cross this
bridge before doing any such modification in the ontology.