Overhaul evaluators

[ttuegel]

Before #1978, I want to overhaul the function and simplification rule evaluators.

Move Kore.Step.Simplification to Kore.Simplify. This is one of the most important sub-domains, so it belongs at the top level of the hierarchy.

We will introduce a new class Evaluator with a single member evaluate with the same signature as BuiltinAndAxiomSimplifier. The latter will be an instance of the class. We will introduce an existential data type SomeEvaluator which can wrap a value of any instance of Evaluator. The Simplifier monad will by modified to hold a map of SomeEvaluator instead of BuiltinAndAxiomSimplifier.

We will move Kore.Step.Function to Kore.Function. definitionEvaluation will become Kore.Function.User.User (user-defined function evaluators) an instance of Evaluator. simplificationEvaluation will become Kore.Simplify.User.User (user-defined simplification evaluators), also a distinct instance of Evaluator. Basically, we will de-functionalize definitionEvaluation and simplificationEvaluation.

[ttuegel]

While we are at it, we should get rid of the big map of evaluators. We should make known function calls whenever possible. The dispatch logic for functions should be:

1. Does the symbol have a hook? If so, match on the hook name and dispatch the appropriate function. (Having a big case with all the hook names sounds silly, but it's quite efficient and also developer-friendly.) The hook evaluators are no longer existentially-quantified types, so dispatching built-ins will be more efficient.
2. If the symbol is not hooked, or the hook did not apply, look up a function evaluator. Okay, here we have to fall back on a big map of evaluators.
3. If no function evaluator applied, look up a simplification evaluator.

We probably won't see much benefit from the known-function dispatching in (1) just because dispatching built-ins isn't a prime cost center for us, but it won't hurt.

This split into three branches should happen in evaluateApplicationFunction.