[Phase 0] Snapshot the Haiku 4.5 triage token-usage baseline before its 30-day artifact retention expires #2102

Story

As a maintainer running the Haiku 5.5 canary comparison,
I want a durable snapshot of the current Haiku 4.5 triage token-usage records taken now,
so that the cost and latency baseline still exists when the canary report compares Haiku 5.5 against it weeks later.

Acceptance Criteria

1. Every token-usage-<run_id> artifact still available from pr-review.yml runs (uploaded with retention-days: 30) is downloaded into one directory, with no record edited or filtered.
2. The directory is re-uploaded as one artifact (or another durable store the maintainer picks) whose retention outlasts the planned canary window. Its name, location, record count and date range are recorded in a comment on the epic.
3. The snapshot holds at least one record whose model matches claude-haiku-4-5* with tier triage. If none exists, the comment says so explicitly.
4. The snapshot is treated as an immutable baseline: later stories read it and never rewrite it.

Tasks / Subtasks

• List and download the live token-usage artifacts (AC: test issue from agent #1)
  • Reuse the artifact-listing approach in .github/workflows/actions-fleet-monitor.yml (it already selects artifacts whose name starts with token-usage-)
• Store the snapshot durably and record where it is (AC: Go-live improvements for PR review agent #2, Optimize review: small-PR and incremental fast paths #4)
• Verify Haiku 4.5 triage coverage in the snapshot (AC: Add @claude delegation, auto-merge, and rebase handling #3)

Dev Notes

• Time-critical: pr-review.yml uploads token-usage-${run_id} with retention-days: 30, so Haiku 4.5 triage data older than 30 days is gone. Do this before anything else in the epic.
• Records are JSONL produced by the engine's token logging (see docs/token-report.md § Real token usage). Copy them byte-for-byte. The canary report ([Phase 2] Canary go/no-go report: compare deep-tier cost, cache-read cost and latency by model from token-usage artifacts #1951) reads this format.
• This is a hands-off operational task because it needs maintainer credentials and a choice of storage. It changes no code.

Project Structure Notes

No repo files change. The output is an artifact plus a comment on the epic.

References

• .github/workflows/pr-review.yml
• .github/workflows/actions-fleet-monitor.yml
• docs/token-report.md
• discussion Enable Haiku 5.5 #2095 (risk: baseline expiry)

Story prepared by the BMAD Scrum Master (Bob) for epic #2101. Status: ready-for-dev.