Experiment with Cython

[gsnedders]

If we can simplify some of the code slightly, we might be able to get some decent performance increase under CPython from using Cython (note that we almost certainly want to keep our codebase pure Python, so this means using Cython's pure-Python mode). It's probably at least worth experimenting here.

At least for html5parser, making decent progress is blocked on #272, and relatedly making html5parser.HTMLParser.phases not a dict (so they can all become struct references).

[gsnedders]

#24 is the other part that would be significant here, I think.