docs(ops): finish self-host rate-limit dashboards and runbooks

[JSONbored]

Parent: #1936

Problem

The self-host dashboard is useful but still has blind spots. Beta operators need an operations runbook that explains queue growth, admission deferrals, rate-limit recovery, AI provider failures, Qdrant behavior, and Orb telemetry status.

Requirements

• Dashboard panels should line up with current metric names and labels.
• Alerts should distinguish backlog caused by local failure from backlog caused by GitHub admission deferral.
• Troubleshooting docs should include exact log event names and PromQL examples.
• Missing/no-data panels should either be wired correctly or documented as optional/profile-dependent.

Deliverables

• Update Grafana dashboard panels for queue, GitHub API budget, cache, admission deferrals, Orb relay/export, provider failures, Qdrant, and HTTP latency.
• Update alert rules for sustained backlog, rate-limit response spikes, admission deferral duration, dead-letter growth, and provider outage.
• Update self-host operations/troubleshooting docs with copy-paste checks.
• Add validation for dashboard JSON / alert rule syntax if available.

Acceptance criteria

• An operator can explain a growing queue from Grafana alone.
• Alerts do not page on expected temporary admission deferrals unless they become sustained/stale.
• No high-cardinality labels are introduced.
• Docs match current metric names.

Expected outcome

Beta users can operate the stack without maintainer SSH access, and we can support incidents from screenshots/metrics instead of guesswork.

[linear-code]

JSO-2028