feat(monitoring): add backend-listen WS connection drop alert

[beastoin]

Summary

Adds a critical alert for backend-listen active WebSocket connections dropping below 10. This catches any backend-listen failure that causes client disconnections, regardless of root cause (e.g., helm empty defaults incident).

Alert specification

Field	Value
UID	tz_backend_listen_ws_drop
Expression	sum(backend_listen_active_ws_connections{job="backend-listen-metrics"}) or vector(0)
Threshold	< 10
Duration	5m
Severity	critical
Dashboard link	Omi Services Overview → Panel 82 ("Backend-listen active WS")
instatus_component	live-transcription

Baseline (3-day)

• Min non-zero: 52
• Average: 218
• Max: 365
• Threshold (10) is ~5% of minimum, giving wide margin before false positives

Expression alignment

Query matches dashboard panel 82 exactly: sum(backend_listen_active_ws_connections{job="backend-listen-metrics"})

Test plan

• JSON syntax validation: traffic-zero.json passes json.load()
• Dashboard annotation: panel 82 resolves in Omi Services Overview
• Expression matches dashboard panel 82
• Deploy via grizzly apply and verify alert evaluates to OK (mon)

🤖 Generated with Claude Code

[beastoin]

Confirmed, this approach is sound. Panel 82 is valid under omi-services-overview.json on dashboard uid 6d3d0897-c5a8-4f7b-b034-a3718c52a63b, and the or vector(0) behavior matches the existing traffic-zero rules where metric absence should page the same as zero traffic. The <10 threshold with for: 5m also lines up with the ops-specified 3-day baseline, so I do not see a remaining blocker here.

---

by AI for @beastoin

[beastoin]

All checkpoints passed — PR ready for merge

Checkpoint	Status
CP0-CP3	✅ (preflight, issue, workspace, exploration)
CP4 CODEx	✅ (2 turns — threshold/vector(0)/panel verified)
CP5-CP6	✅ (implemented, PR created)
CP7 Review	✅ (PR_APPROVED_LGTM — JSON valid, PromQL correct, panel 82 resolves)
CP8 Tests	✅ (TESTS_APPROVED — config validation suite)
CP9A/B	✅ (JSON + annotation verification)

Awaiting manager merge approval.

by AI for @beastoin

[beastoin]

lgtm