Add support for inductively defined functions

[stevenraphael]

Following substantial discussion with Andrew Adams and Alex Reinking, I have implemented a first draft of their idea of inductively defined functions. A full description is in src/Inductive.h. This pull includes correctness and error tests. Please suggest any additional changes I need to make. The basic idea is to have pure functions that are defined with self-references, e.g. f(x) = select(x <= 0, input(0), input(x) + f(x - 1)), which allows for optimizations that are currently impossible to do with reduction domains.

[zvookin]

Given the constraints on the definition, I'm wondering if this should be its own syntax. E.g. f(x) = inductive(cond, expr1, expr2) Could also try to put it all in the construction, similar to how RDom is done, but given the arguments need to be specified, function definition syntax seems better.

Main goal is to avoid confusion around the lack of generality in the definition. If it is envisioned that this will become more general, the current design might be the right way to go. (And by confusion, I mean that this makes it look like normal programming language recursion expectations are supported.)

[abadams]

The goal is definitely to become more general over time. The thinking was that if we fail to solve for a base case, we could tell the user to write things in a particular form that is known to work, but first we'd try to handle any recursive RHS.

[alexreinking]

We are likely to merge #8877 before this, so consider using the new APIs to simplify visitors/mutators here. I see two here: RecursiveHelper in is_inductive, and maybe BaseCaseSolver as it has just one visit overload.