Add FastText-based semantic neighbor search rather than relying on relatedwords.org

[jonadsimon]

Write logic to fetch semantic neighbors from FastText offline.

Need to add eye-balled constraints on semantic relatedness to ensure that enough words are generated so we can populate large puzzles, but not so many that the meanings become disassociated or the strings are misspelled garbage.

Enforcement variables include:

1. # of neighbors
2. distance of neighbors
3. frequency of neighbors within the training set (i.e. discard overly-rare words, since they're probably typos)

[jonadsimon]

Decided to stick with relatedwords.org after realizing how well developed it is. Wrote code to ping it with a GET request instead of copy/pasting. Will continue using it in this manner until otherwise noted.