Improve entity similarity endpoint

[cthoyt]

This PR improves the entity similarity endpoint in the following ways:

1. Switches from returning the cosine distance to a normalized version of the cosine similarity, such that similar entities get 1.0, dissimilar entities get 0.0, and sort of similar entities get 0.5 (average)
2. Rename the data model returned by the entity similarity endpoint and update the key inside it from distance to similarity
3. Improve documentation inside the response model and fields
4. Enable returning self-comparisons, which should always result in 1.0