healthCheckCacheTtlMs

How long (in milliseconds) a /kandra/health result is cached before the next request triggers a fresh SELECT release_version FROM system.local against the cluster (GH-36). Without this, a probe storm (misconfigured monitoring, or deliberate abuse if the route is reachable beyond an orchestrator's internal network) translates 1:1 into real cluster queries. Set to 0 to disable caching and query on every request.