Cluster-availability probe

Returns 200 only when the node is part of a healthy cluster (Raft state healthy and a leader elected). Disk and clock skew gate writes (see admission), not cluster readiness. Returns 503 otherwise. Use this for external monitoring or any client that needs to wait until the node can actually serve cluster-dependent traffic. Kubernetes readiness probes should target /readyz instead — this endpoint MUST NOT gate StatefulSet ordering, otherwise the cluster cannot recover from a full-kill cold start.