Translate between EGraph types

[sgpthomas]

Addresses issue #305.

I use some pretty complicated trait bounds in the EGraph::map function to handle converting between analyses. If the analysis stays the same (EGraph<L, T> -> EGraph<L2, T>) these are automatically implemented which makes it more convenient to use in what I imagine is the common case. However, let me know if you think the conversion should be explicitly handled with a passed in closure.

[mwillsey]

Here's an alternative idea that might ease some of the trait bounds, and it would make things a little more flexbile than requiring From. Make LangMapper a trait or something:

trait LangMapper<L, A> {
  type L2: Language;
  type A2: Analysis<Self::L2>;
  fn map_node(&self, node: L) -> Self::L2;
  fn map_analysis(&self, ...)  {
    // should maybe be able to provide a default impl?
  }
  fn map_egraph(&self, egraph: EGraph<L, A>) -> EGraph<L2, A2>
}

If you wanted to go crazy, those methods could all take in the src and dst egraphs? What do you think?

[sgpthomas]

I liked that idea and implemented it. I couldn't figure out a nice way to provide default implementations in the LanguageMapper trait, so I wrote a SimpleLanguageMapper struct that provides a LanguageMapper implementation when there are the necessary From<..> impls. Those type constraints are still pretty gnarly, but they aren't on the default interface. Let me know what you think of that.

[mwillsey]

Looks good to me! Thanks for the contribution. Seems like some doctests are failing. Could you patch that up? Try cargo test --doc. Btw make test does everything the CI does.

[sgpthomas]

Okay, should be good to go

[mwillsey]

Thanks for the contribution!