add some failing tests that we need to fix

add zero width character pattern in utf8 module

improve url tokenizer to handle domain hiding through username@

lift domains before sampling so that domains can be captured

add system for uncertainty weighting

remove legacy alternate classifiers that never really worked

add token weight multipliers for boosting

comment

woops, missing new utf8 module

change where unaccent takes place to clean up before splitting

add cjk splitting support

rewrite token merging to try to keep ordering better

fixing up tokenization merging a bit, add domain_tokens_first option

add filter tokens callback to the train_text method on classifier

add a train_text method to the classifier to simplify the bundling of the tokenizer and training