Detects screenshots and classifies them on-device (receipt, error, chat, ticket, map, product, code), extracts entities and suggests actions.