Triage 7 stale winston/ PRs with unresolved cursor review threads

Per HEARTBEAT.md Step 1a (added 2026-06-12), every heartbeat surfaces unresolved review threads on open winston/ PRs and MUST act on them. As of 2026-06-13 14:42 UTC, pr-watch-unresolved-feedback.sh flags 7 PRs with unresolved cursor-bot review threads, all 4–6 weeks old: - PR #10566 — 2 unresolved (3CE Customer status): non-transactional read; duplicate enum export name - PR #10565 — 2 unresolved (BullMQ shutdown guard): missing guard in onConnectionWorkerFailed; missing guard on QueueEvents failed - PR #9996 — 1 unresolved (Tesla command.executed): unsatisfied peer dependency after event-bus version bump - PR #9936 — 2 unresolved (Enphase onlineStatus): status updates can be skipped; cached gateway data stales status - PR #9934 — 1 unresolved (Sensibo onlineStatus): missing stale-reading detection - PR #9928 — 1 unresolved (FranklinWH onlineStatus): stale-reading detection not implemented - PR #9584 — 1 unresolved (ALL-572 ManufacturerCredentials): duplicated maskClientId utility Acceptance: 1. For each unresolved thread, classify: VALID (cursor is right — push a fix commit), STALE/SUPERSEDED (file/code no longer matches concern — resolve thread with brief explanation), or NEEDS-RICHIE (architectural decision — leave with summarized question for Monday DM). 2. For VALID concerns, spawn one fix sub-agent per PR with targeted scope (per cursor's file + body). 3. For STALE/SUPERSEDED concerns, resolve the GitHub review thread via API with a one-line explanation. 4. Produce a single summary update on this fleet-task with per-PR classification and outcome. Context: A previous backlog DM at 04:14 UTC (~10h ago) asking Richie a/b/c/d on 16 stale winston/ PRs is still unanswered. Richie engaged with me at ~14:00 UTC on workspace health (separate topic). The cadence-limiter (HEARTBEAT.md Step 2b) suggests deferring non-urgent decision DMs, but addressing concrete cursor concerns is engineering work, not a decision DM.