Apply optimizations all across the board.

[edponce]

There are places for improving runtime performance:

• Use local variables as proxy for class variables
• Tokenizers split vs regex vs merged sentencize/tokenizer
• Compile regexes
• Async processing
• Streaming processing
• Compress serialized data in stores
• In Elasticsearch commands, predefine dicts with defaults instead of creating dict for every search.

[edponce]

Consider LRU cache from functools, https://docs.python.org/3/library/functools.html#functools.lru_cache

[edponce]

Try to improve _make_feature_set() in ngrams. Solve the unique-ordering of features, that is do not append integers.