Need improvements to feature hashed vectorization for speed and flexibility

Need to have a text vectorizer that handles a lucene analyzer.

Also need byte[] short-cuts to avoid converting to a string and then back to bytes.