Improve classify_format (with a parser?)

[brendanlong]

classify_format uses a crazy regex and several String.contains checks. We should replace it with a Menhir parser, which would be substantially faster and probably more readable.

Note: I don't know how much the performance of this function actually matters right now, but this is an improvement I'd like to do because it's both faster and cleaner.

[brendanlong]

Menhir is also just an example. Some other parser format might be reasonable too (Angstrom? something custom?). The main thing is that I want the code to be relatively straightforward (preferably functional), and preferably not have to make several passes of each string.