Minor metrics gaps: Service port, ServiceMonitor interval, status.endpoints.metrics

[bmorton]

Low-priority cleanups found while wiring Prometheus scraping for temporal-benchmarks. None block usage (metrics on :9090 work by default), but they surprise users who customize metrics:

1. spec.metrics.port ignored by the Service. buildMetrics() uses the configured port in listenAddress, but BuildHeadlessService/containerPorts hardcode metricsPort = 9090. Setting spec.metrics.port to a non-default value makes Temporal listen on the new port while the Service still exposes 9090, breaking ServiceMonitor scraping.

2. spec.metrics.serviceMonitor.interval ignored. BuildServiceMonitor hardcodes interval: 30s and never reads the CRD field.

3. status.endpoints.metrics never populated. The field is declared on EndpointsStatus but the reconciler never writes it.

Suggested fixes: thread cluster.Spec.Metrics.Port into the container/Service ports; use cluster.Spec.Metrics.ServiceMonitor.Interval (fallback 30s) in BuildServiceMonitor; populate status.endpoints.metrics during monitoring reconciliation.