Operators, Guardrails, and Trajectory-Level Integrity in Human–AI Interaction: A Structural Framework for Conversational Possibility Space | Synapse