Create PZ Index Class which Can Be Used by Retrieve

[mdr223]

Right now we do not have a strong abstraction in place for indices which are used by the Retrieve operator.

As a result, we require the user to do some heavy lifting in writing a search_func() which uses their index and returns results to the RetreiveOp.

In an effort to mitigate this heavy lifting -- and to make it easier for us to program against a standardized class -- this issue aims to add a BaseIndex class, along with sub-classes for ChromaIndex and RagatouilleIndex.

These indices will expose:

• a __str__ function which can replace the need for index_helpers.py
• a query function, which takes a query: str | list[str] and a results_per_query: int and returns a list | list[list] with the top results_per_query results for each query.

This issue will also standardize the semantics of the search_func. If the user's search function returns a list[str], then RetrieveOp will take the top-k elements from that list. If the user's search function returns a list[list[str]] then RetrieveOp will take the top-k elements from each sub-list.

[mdr223]

Based on conversation w/ @sivaprasadsudhir, we will rethink the interface for Retrieve for the longer-term -- but for now we have a short-term solution.

[mdr223]

A couple of notes:

• RAGPretrainedModel can only perform search w/string input (not embedding(s))
• Chromadb.Collection may not have the correct model name if the user creates the index w/out specifying the embedding_function (which you technically don't need to do if you are only querying the index)