Fix a bug in mk_superclasses_of

This bug meant that we were less eager about expanding
tuple superclasses than we should have been; i.e. we stopped
too soon.  That's not fatal, beause we expand more superclasses
later, but it's less efficient.