M16.7 Service health: checks, dependencies, incidents #70

Part of #63. Needs M16.4 (alert delivery and the uptime runner).

yuva-api

• monitor_services (name and environment, optional API property, push or pull with a health URL and interval, silence timeout), monitor_dependencies (kind, critical, latency), monitor_instances (gone after 24 h of silence); made by the first report, a member, or the first record or span that names the service.
• POST /v1/monitoring/health (analytics:write) for pushed reports; pulled reports through the uptime runner and its address checks, reading application/health+json, Spring Boot Actuator or the status code alone; messages shown to members only.
• State per service and dependency (healthy, degraded, down, unknown) from reports, critical and non-critical dependencies, the silence timeout, linked uptime checks and the linked property's 5xx rate and p95 over 5 minutes; recovery after two healthy evaluations; the monitor_states job every 30 s; monitor_state_changes (400 days) and monitor_health_hours (90 days).
• monitor_incidents and monitor_incident_updates: one automatic incident per service turning degraded or down, resolved after 5 minutes healthy; manual incidents and maintenance windows (no alerts or automatic incidents inside one); monitor.incident stored event and webhook.
• State-change alert rules of M16.4 on services and dependencies; the health dataset; MCP tools for current state and incidents.

yuva-widget, yuva-api (SDKs)

• sdk/go/health and createHealth in useyuva/server (with the srvx, h3, Nitro, Hono, Express, Fastify and Next.js wiring of M16.2): register checks with timeouts and criticality, serve application/health+json, push reports on an interval.

yuva-panel

• Monitoring section: services overview with state, dependencies and incidents; a service's page with its checks, instances and history; incidents with updates and maintenance; a health panel type for dashboards (M16.6). en and tr.

yuva-docs

• Health checks per stack, pushed versus pulled, how state is derived; en, tr, de.

Done when

Stopping a critical dependency of a sample service turns the dependency down and the service down within the configured time (a non-critical one degraded), opens one incident and one alert, and both close when it comes back; a service that stops reporting turns down after its timeout; a pulled Actuator endpoint and a pushed Go report give the same states.