Faster tokenizing and consistent use of token lists

[fidian]

The current state of the code has about three different token-ish implementations. This should be consolidated down to one. The tokenizer is also really slow on lengthy CSS strings and should be made faster somehow.

[fidian]

Progress is being made. I'm moving the code over to use the Complexion library. It was written with better speed testing and tuning. Once that happens, PrettyCSS will need to consume an alternate token stream and then I can standardize how these streams are passed around.