[leecyang]

Production Worker startup reconciles existing IM channels before starting the agent runner. Reconciliation constructs local LingxiOS context with the active API role; runner admission then rejects that context, closes PostgreSQL and leaves maintenance timers alive, so the process never exposes its readiness listener.

Select the passive Worker role before storage initialization and reconciliation, and reuse that ownership check at runner admission. Reconciliation still completes before task claiming; active API contexts remain rejected in production. Published runtime, schema, authorization and model settings are unchanged.

Validation: an integration regression using isolated PostgreSQL first reproduced the wrong control-listener ownership after successful channel reconciliation (run 39027258962). The fix is checked through the normal server/typecheck and integration CI jobs. Production recovery will use a separately reviewed immutable Worker image pin and Komodo service updates.