Drain terminal usage after expected context-append close errors

When an OpenAI GPT-Live client closes while context appends are still pending, the provider can report context_injection_incomplete before session.closed. Classify that exact error in the closing state as an unconfirmed context delivery, then keep receiving the terminal event and final usage. Treating every error as a fatal transport failure can discard a valid finalization record. Keep other errors fatal and bound the drain wait; this exception does not prove the appended speech was delivered or heard.

Applies when: A GPT-Live receiver closes while a commentary, thinking, or instruction append remains pending. Keywords: OpenAI GPT-Live WebSocket context append close finalization usage receiver