Finish test result

Finish a test result and start processing and scoring in the background. The status flips to completed once scoring lands — poll the test result to observe it.

For test results with completionMode: manual: after next-message returns end_chat: true, attach any remaining data to the conversation (metrics, tool calls, events), then call this endpoint. Finish all conversation writes first — inserting and updating messages is rejected (409) from this call onwards.

Calling this on a result that is already being processed is rejected (409), so a retry never scores the same conversation twice. Returns the result's id and status.