No Chunks, No Embeddings: OpenAI’s Index‑Free Long RAG