Start a patient conversation or send a patient reply

Requires GP scope. Existing conversations/templates require resourceId and their latest expectedVersion. New conversations require patientId. Commands are validated by role. Mutations are attributed to the authenticated team; incoming/outgoing direction identifies the simulated role. Patient responses omit private practice information. Practice messages are queued until a delivery command marks them delivered. configure_auto_reply sets an ordered script for future successful deliveries: each delivered outgoing message consumes one step and schedules a reply after delayMinutes of simulation time. Configure before delivery, then POST /api/clock with paused:true and advanceMinutes. Reconfiguration resets the script and cancels pending replies; empty steps disable it. Manual patient replies and completing a conversation cancel pending replies without reusing consumed steps. Read the latest conversation version after clock changes. Existing generic messaging_action commands remain supported.