recall-rag 0.9.5

Trustworthy retrieval for an AI agent’s own memory — verdict, confidence and provenance per hit, or an explicit abstention