The methods here are sort of weird. It would make more sense to just
have newSWFMorph(shape1, shape2);, but as things are now, shape2 needs
to know that it's the second part of a morph. (This, because it starts
writing its output as soon as it gets drawing commands- if it kept its
own description of its shapes and wrote on completion this and some
other things would be much easier.)