Conversion to CPS

[William103]

Implement #17. I'm currently almost done, but things are still very rough around the edges.

TODOS in no particular order:

• Actually use compile_match_tree and convert_decision_tree in convert_expr
• Use choice_defs from hir program to get a list of all variants for each choice type (i.e. implement resolve_path)
• Rework match compiler to use an arena, or at the very least Rc's, rather than just cloning everything
• Ditto for CPS converter
• Test CPS converter on piecewise closures
• Investigate potential bugs in match converter (it generates correct decision trees, but also seems to generate several unnecessary comparisons)
• Generally clean up, optimize, and document things (both algorithms are rather obtuse and mysterious)

Eventually, probably in a different PR, I'd like to write a new interpreter that interprets the CPS code.

[William103]

... can you just make everything methods on a struct?

That was my first attempt. I'm translating some SML code, so it gets wild with all the references. The callbacks everywhere would end up needing to take a &mut self as well as convert_record and its callbacks, and that's too many mutable references for the borrow checker to be happy. It could maybe be possible with a bunch of Rc<RefCell<self>> but that sounds horrible. I'm sure there's a way to do the whole algorithm more imperatively with a struct but idk how; I barely understand the functional version.