[core] Mutator result caching

After implementing mutator caching for the third time in one week, I'm starting to think this is something the core should do for us.

Essentially:
- A Mutator should be able to declare if it is "pure."
- The result of applying a "pure" Mutator to a type T multiple times (by object identity), should return an object-identical result if and only if `T` is _finished_.
- If `T` is not finished, mutator callers will be responsible for caching mutation results if they want, or we could provide another mechanism to select/filter types that should be cached in that circumstance.

Of course, nothing in the language will stop someone from mutating finished types, but please don't.