M16.4 API alerts, uptime checks, reports, MCP #67

Part of #63. Needs M16.3.

yuva-api

• monitor_alert_rules per property, route, service or check: error rate, client error rate, p95, p99 or Apdex over a 5–60 minute window with a minimum of requests; spike and drop against the same 15 minutes of the 7 days before; a route going silent; certificate expiry. The monitor_alerts job every minute; one open alert per rule and key that fires once and resolves once (10% margin over a whole window). Delivery: monitor.alert stored event and webhook, an annotation, e-mail and Web Push to the rule's recipients. A default 5xx rule on every new API property.
• Uptime checks (monitor_checks): HTTP(S) GET or HEAD, expected statuses, keyword, timeout, interval (at least YUVA_MONITOR_MIN_INTERVAL), failures before down, certificate warning, redirects; monitor_checks_due every 10 s and one monitor_check job per run on a new monitoring queue (YUVA_MONITOR_WORKERS); requests through the webhook client's dialer (private, loopback, link-local, CGNAT, multicast and reserved addresses refused, every redirect hop checked, YUVA_MONITOR_ALLOW_PRIVATE for self-hosted, metadata always refused); a retry after 20 s before a failure counts; minutes where most of the server's checks fail recorded unknown; monitor_check_results (day partitions, 30 days) and monitor_check_days; caps YUVA_MONITOR_CHECKS and 10 checks per host.
• The uptime dataset in the query operation (up, uptime, response-time percentiles, ttfb, certificate days left).
• Weekly and monthly API reports through analytics_subscriptions and the analytics_reports job with an API template.
• Support in aggregate: conversations per local day of the linked inbox next to errors, by time only; no link from a request to a person.
• MCP tools: slow and failing routes, a route's history, a consumer's errors, alert state, uptime checks; the new datasets in the query tool; setup_api_monitoring prompt; get_setup_state entry.

yuva-panel

• Alert rules and their history, uptime checks with results and timings, report settings. en and tr.

yuva-docs

• Alerts, uptime checks (what the server will and will not call), reports and the MCP tools; en, tr, de.

Done when

A route pushed over its error-rate threshold raises an alert once and resolves once, by e-mail, push and webhook; an uptime check against a stopped endpoint alerts and recovers; a check aimed at a private, loopback or metadata address is refused when saved and when run; the MCP tools answer from the same numbers as the panel.