Extend RAG with RAG Fusion and Rewarding-LLM

[scalabreseGD]

Even thinking that refactor the codebase to work with Knowledge Graph might be too big of change, I believe that implement RAG fusion for Retrieve-Generate makes more sense in terms of accuracy.
Also, There's a need to implement multi-LLm solution where one model can be used as rewarder.
I leave here the link to the paper https://arxiv.org/pdf/2402.03367