Resolve bindings before the denotation phase

[robintown]

Currently bindings are resolved during the denotation phase. This is turning out to be inflexible and requires some hacky workarounds to make binding sites work how we want them to. We could introduce a new phase which assigns indices to all bound structures, or maybe we could even do this as part of the deep structure recovery phase.

[robintown]

Won't fix; in an effectful expression model like the one started on the branch effects, resolving bindings in the denotation phase is both the most natural solution and also a very flexible one.