Refactor Signature to work with arity, HM types, etc.

[joshrule]

We might be able to add a trait to Signature such that it assumes that each atom has a particular label that is used to ensure validity. This could include arity, Hindley-Milner types, and perhaps other sorts of validity constraints. It would also make it possible to relocate TRS from the programinduction crate to term_rewriting and could significantly reduce the complexity of that code.