`ElasticsearchInferenceHybridRetriever` (dense + BM25)

[davidsbatista]

• Add a new retrieval method (e.g. _hybrid_retrieval) that combines dense and BM25 in one request, and optionally merges scores (e.g. RRF). The method should accept either a query vector (from Python) or query text + optional inference_id for the dense branch.
• New retriever component: calls the document store’s hybrid retrieval method; no changes to existing retrievers.
• ES: RRF (if used).

[sjrl]

Just out of curiosity why not make a super component like we did for Open Search?

[davidsbatista]

We can do that one, but as far as I understand ES also offers the same retrieval method as an inference service - the idea of this issue is for that ES inference service.

I will update the name of the component so that it's explicit.