Add reranker

[carlosmada22]

A reranker is a more sophisticated model that takes the smaller list of documents from the retriever and re-scores them based on their actual relevance to the user's specific query

The reranker model (often a "cross-encoder") looks at the user's query and each document at the same time, which allows it to make a much more accurate judgment than the vector search alone.

The new workflow will be:
User Query -> Embed -> Retrieve Top 20 -> Rerank to find the actual Top 5 -> Generate Answer

This adds a small amount of processing time but dramatically increases the chance that the context given to the LLM is the best possible, leading to much better answers.