Improving RAG Retrieval via Propositional Content Extraction: a Speech Act Theory Approach