Implement a customer review sentiment classifier under prebuilt [resolves #192]

[dafajon]

bblock/doc.py:

• Implement a fix for returning zero vector instead of nan when the denominator/normalizing_factor is 0.

prebuilt/data:

• gzipped training data. Evaluation data is exclusive not included.

prebuilt/customer_sentiment.py

• Implement empty_model, build and load methods.
• Implement class mappings.
• Parameters for vectorizer and models are chosen in 3 fold cv. Optimizer code is exclusive not included.
• Save model with version. Use the version number for loading by adding version argument.

Update prebuilt/README.mdwith usage and evaluation accuracy.

[husnusensoy]

1. Built on 8K instances is not a serious model to be included into model
2. None of the code/data naming, layout standards is followed.

Thanks for your effort but kindly rejecting this PR