Adding in a rockset retriever class

[danpechi]

• the input can be finicky; it works with something like: str(openai_client.embeddings.create(model="text-embedding-ada-002", input=[question]).data[0].embedding)

• this has some inflexibility wrt underlying schema, ie all results are coming from a singular vector db collection

• note: rockset doesn't have a 'retrieve' function, it just leverages cosine_sim

• this is mostly meant for singular questions, and not batch inference

[arnavsinghvi11]

comment to be addressed? This can be a dspy.Prediction(passages=...).

[arnavsinghvi11]

Thanks @danpechi for the PR!

Couple of changes needed to merge:

1. To support the import message here, can you add rockset as a dependency to the pyproject.toml? Feel free to follow how it's done for the other retrievers and then add rockset as a supported RM in the README (under Installation).

2. Please run ruff check . --fix-only and push to pass the failing ruff test.

It would also be great if you could add documentation for the Rockset RM to provide context as done for the RMs.