use nltk.classify.apply_features for memory efficiency

[eguven]

Using this instead of building the entire list freed up significant amount of memory with large datasets. Source: nltk doc