pbi-escalation-handlerlisted

## Inputs - Notification from Developer (Agent Teams) with PBI id and `escalation_reason`. Backlog `items[].status` for the PBI is `escalated`. - `.scrum/pbi/<pbi-id>/state.json` (`escalation_reason`, round counters, per-stage `*_status`) - Latest review files: `.scrum/pbi/<pbi-id>/{design,impl,ut}/review-r{last}.md` - `.scrum/pbi/<pbi-id>/metrics/*.json` ## Outputs - SM judgment recorded at `.scrum/pbi/<pbi-id>/escalation-resolution.md` (audit trail) - backlog.json `items[].status` updated via `update-backlog-status.sh`: - **retry** → `in_progress_design` (round counters, per-stage `*_status` flags, and `merge_failure_count` reset on `state.json`; worktree preserved). **kind=docs PBIs retry to `in_progress_impl` instead** — design was never run — with `design_status` and `coverage_status` reset to `skipped` (not `pending`) and `ut_status` left at `pending`. See § Steps step 4. - **hold** / **human-escalate** → stays at `escalated` (until the blocking condition clears, at which point SM moves it to `in_progress_design` to resume; worktree preserved for inspection) - **block on external dependency** → `blocked` (SM-only status; later transitioned back to `in_progress_design` when the external factor clears; worktree preserved) - **abandon** → `cancelled` (terminal); SM calls `cleanup-pbi-worktree.sh` to remove the worktree + `pbi/<id>` branch. The audit trail lives in `escalation-resolution.md`, not in a ling