Why not redesign jacobian(Ideal) so it returns the submodule of the
differentials of the ring that actually depends just on the ideal?  That would
more mathematically interesting than what it does now, and the former behavior
could be retrieved by replacing jacobian I with jacobian gens I.
