Add Regexp tokenizer

[rth]

This adds a Regexp tokenizer, which is, for instance, the default tokenizer in scikit-learn.