Add resilience metrics, dashboards, and alerts

[osvaldoandrade]

Parent epic: #574

Problem

Operators need direct signals for the states that caused the recent incident.

Acceptance

• Metrics cover quorum, leader, Raft term, applied index, identity epoch, duplicate identity, probe state, repair action count, and shutdown phase.
• Alerts cover quorum loss/degradation, duplicate identity, stuck terminating pod, NodeNotReady, failed repair, and long non-ready duration.
• Dashboard shows cluster health, node distribution, repair state, and traffic-serving status.