upstream: no endpoint to query gateway records by correlation ID (or session ID)

[amirkhan-ak-sf]

Part of #611.

Summary

Omni Gateway records every governed call by correlation ID in its analytics, but there's no endpoint to query those records by correlation ID (or by session ID). If there were, run and session history (models served, refusals, usage) could be rebuilt from the gateway itself rather than a client-side store. That would be the most governed form of "memory about what happened", with a single source of truth.

Request to the Omni Gateway team

• GET records for a correlation ID (and optionally a session ID): per call, the timestamp, served model and provider, routing and fallback, policy outcome, token usage, and request ID. No prompt or response content.
• Scoped to the caller's own client ID; an agent can only read its own records.
• Freshness: near real time is enough for human-in-the-loop hand-off, but record the expected lag.

Why it matters to DDK

• It lets the run ledger (telemetry: run ledger — donkey.run() keeps every call, refusal, usage and cache/routing outcome for the run #613) and session scope (memory: session scope — link runs by session ID, backed by the chosen store #620) survive a process restart without DDK storing anything.
• It turns human-in-the-loop resume (hitl: one vocabulary for pause-and-ask-a-human #211) and cost questions (telemetry: a run-scoped rollup answering "what did this task cost" #317) into gateway-authoritative answers.
• It's the same shape of gap as upstream: no budget-query endpoint — remaining budget is only visible in-band #248 (budget only visible in-band).

Related

#248, #317, #211, #613, #620