Feat/elo

[xpbowler]

MR: Elo Leaderboard feature

Uses an implementation of the Elo rating system: https://en.wikipedia.org/wiki/Elo_rating_system

We have 3 leaderboards:

1. LLM: generation LLM's (e.g.,GPT-4o, command-r)
2. Retrieve: the Retrieve pipeline - retriever model + reranker model (e.g., BM25+rank_zephyr)
3. RAG: the whole RAG pipeline - retriever model + reranker model + LLM (e.g., BM25+rank_zephyr+GPT-4o)

A battle is eligible for updating a leaderboard following the conditions stated below:

• LLM — all parameters in pipelines are the same except for llm
• Retrieve — all parameters in pipelines are the same except for retriever and/or reranker
• RAG — the llm's are not equal and either the retriever or reranker's are not equal