Reimplementing regex grammars with finite state machines

[hudson-ai]

This is half "Show and Tell" and half "Idea".

I re-implemented regex grammars using finite state machines (which are created from regex patterns using the interegular library):

https://github.com/hudson-ai/pastafarian

This is not a PR to guidance for a few reasons:

• I don't know if this approach adds enough features to warrant changing the current implementation.
• Using as a drop-in replacement for guidance's implementation causes some tests to fail due to an infinite recursion when regex grammars are wrapped with token_limit. I believe I can solve this, but then I hit another problem with token limiting:
  Calling token_limit on recursive grammars fails to limit tokens #676

@Harsha-Nori, @slundberg, @riedgar-ms you may find this interesting :)