Implement cosine similarity example

[Trogluddite]

This script will:

1. query SOLR
2. build the bayesian network from the resulting docs
3. generate markov chains
4. calculate cosine similarity between generated text and a static match string
5. select either first match that's 98% similar or best match after 500 rounds

this, or a similar method, will allow us to start measuring quality of generated text, which opens two paths:

1. 'searching' a generated text automatically
2. automatic retraining