Add optimizations for maximum entropy tokenization

[nyxtom]

Depending on the version of opennlp and whichever fork you use off of it (language of choice), you can extend tokenization to support certain pre-optimizations such as: preserving hashtags, urls, @mentions, email addresses, emoticons..etc. It would be nice if chalk supported these kind of extensive features.

[jasonbaldridge]

Sorry for the delay on this. I'll look into it as I refactor things. Any help from others welcome!