Reconcile an accountable as-of incident handover

Repair handover.sql. Return exactly incident_id,owner,observed_state,unverified_tasks,overdue_tasks,next_action. Preserve EVERY unique incident declared at/before observation second 1000, including NULL owners. Collapse identical imported rows. Select the latest state_event for that incident with declared_second<=observed<=1000, tie-breaking by lexically greatest event_id. No event means DECLARED; OPEN,CONTAINED,RECOVERED,CLOSED are supplied observations. A later OPEN after CLOSED is the actual selected observation. Never admit pre-declaration or future events. Current verification_tasks are supplied as-of records, not a history to rank. Count DISTINCT required=1 task_id values. A required task is verified only if state='verified' and verified_second is non-NULL within declared_second..1000 inclusive. All other required tasks are unverified. Nonrequired checks do not change readiness. Overdue counts distinct unverified required tasks with known due_second<1000; exactly 1000 is still within its inclusive deadline. NULL due dates remain unknown, not automatically overdue or verified. Local next_action precedence: NULL owner -> ASSIGN_OWNER; zero required tasks -> DEFINE_VERIFICATION; selected CLOSED with unverified tasks -> REVIEW_CLOSURE_EVIDENCE; otherwise any overdue -> ESCALATE_BLOCKER; otherwise any unverified -> TRACK_VERIFICATION; otherwise RECOVERED or CLOSED -> REQUEST_OWNER_REVIEW; all remaining states -> CONTINUE_RESPONSE. Retain raw observed_state even when closure evidence is deficient. No next_action executes a response or grants approval. Window-ranked status or anti-dominated event selection is valid. Inspect actual selected status, evidence counts and unresolved owner/deadline; save and repair without Reset.