Change similarity.py to use gensim's Similarity class

[JohnHBrock]

This will ensure there aren't out-of-memory errors and should also make some of the logic simpler.