feat(jit): implement a bytecode VM

[TogarashiPepper]

Currently the interpreter is very slow, to the point of it being really bad for anything that isn't simple. Computing fib(25) (with the recursive approach) takes ~52.5ms with node.js but in the koan interpreter it takes ~188.2ms. Hopefully a bytecode VM will help

TODO List:

• >90% test coverage on all files, 100% function coverage
• Reuse ExprPool in the REPL (have a global pool?)
• Make TokenStream own its ExprPool
• Implement variable semantics
  • Figure out final semantics of how values will work, refcounting? a full tracing gc?
  • simply leak the memory? (probably not)
  • https://en.m.wikipedia.org/wiki/Boehm_garbage_collector
• Implement function definitions
• Implement function calls
• use type checker and make test cases for it
• make arrays homogenous
• make arrays SIMD'd
• improve how value::{add, sub, mul, etc} are implemented (since theyre super messy rn)

[TogarashiPepper]

Pivoting from VM to llvm JIT because I think it's feasible and we'd get all the llvm optimizations "for free"

[TogarashiPepper]

update: it was very much not "for free"