added alternative embedding models for sentence transformers and openai

[jamescalam]

Addressing #99. Added ability to choose embedding model via the config.yaml. Also added OpenAI API as option (alongside Sentence Transformers). We can do:

models:
  - type: main
    engine: openai
    model: text-davinci-003
  - type: embedding
    engine: openai
    model: text-embedding-ada-002

and

models:
  - type: main
    engine: openai
    model: text-davinci-003
  - type: embedding
    engine: SentenceTransformers
    model: all-MiniLM-L6-v2

I understand there is also some work ongoing in this space, but in a private repo, naturally, I can't see this, so I apologize if any of this conflicts with or overlaps. Let me know if any of this can be of use!

Thanks

[drazvan]

Thanks for this @jamescalam! I'd like to merge this in the next few days. Let me check against the internal work, but I think we'll first merge this and then apply the rest of the changes.

A couple points:

1. After merge, we'll likely change the interface to be async. Otherwise, in a server scenario whenever an embedding is computed using an endpoint, like with OpenAI, everything is blocked.
2. We should probably add a register_embedding_provider method as well so that a user can register a custom one from their config.

Nice work! 👍