Implement CharNGram Based HashVectorizer

[dafajon]

• Receive ngram_range.
• Calculate n_grams within a Doc object's lowerized tokens.
• Send (n_gram, str) pair iterator to FeatureHasher.
• Implement tests with tokenizer and ngram_range configurations.

[husnusensoy]

Please do add a model which got improved by the use of this Transformer

[dafajon]

I improved telco_sentiment with Model test accuracy (accuracy): 0.7040786809372288. Moving on to other prebuilts that use hash.

[husnusensoy]

Please do add that into your pull request. We can cherrypick if needed