Retrieve retained run events

Returns public Pi and Salambo events in run sequence order. Requires run:read. Event payloads are read from the encrypted S2 stream.

Events are technical history with their own retention, stored apart from the run record. The durable record of a turn, including its status, output, error and agent_version, is the run object from getRun, which stays readable when events are not.

History is a rolling window. Each run keeps the retention period its event streams were provisioned with (14 days by default), and events expire one by one as they age past it, so older events can be gone while newer ones remain. The window is applied here: an event older than the retention period is not returned, even if it is still stored. The period is returned as history.retention_days, and an entitlement change applies to runs created afterwards, not to existing ones.

A read without after reports what it found in history.status: complete when the events begin at the run's first (or the selected turn's first), partial when older events expired and only a retained suffix is returned, expired when events were recorded but every one has expired, and empty when no event has ever been recorded. A page never presents a retained suffix as the whole history. A read with after continues a page and reports no status.

A cursor whose event has expired returns 410 with code event_cursor_expired. 503 with code event_boundary_pending means the first event of an admitted turn is not recorded yet, and 503 with code event_history_unavailable means the history service could not be read, or a stream that held events is missing. Both are retryable, and neither is ever reported as empty or expired.