RAG: comparison by supported models, partial answer for 1 known model, stronger prompts for context aggregation

tests