Duplicate time series causes 503 on /metrics endpoint

Problem: StackDriver still has issues dealing with duplicate time series, which results in an internal error that bubbles up on the /metrics endpoint with a 503 response code.

Expected: Duplicate time series are properly handled internally, and requests to /metrics will always return metrics in valid Prometheus format.

Sample error returned to Prometheus agent:

An error has occurred while serving metrics:\n\n[from Gatherer #2] collected metric "stackdriver_vpn_tunnel_vpn_googleapis_com_vpn_tunnel_gateway_ip_version" { label:{name:"gateway_ip_version" value:"ipv4"} label:{name:"project_id" value:"foobar-network"} label:{name:"region" value:"europe-west1"} label:{name:"tunnel_id" value:"REGION:1100/PROJECT:<REDACTED>/VPN_TUNNEL:<REDACTED>"} label:{name:"unit" value:""} gauge:{value:1} timestamp_ms:1753891144410} was collected before with the same name and label values

Start with collectors/monitoring_collector.go, especially lines 262-274, and trace how duplicate StackDriver time series reach the /metrics handler. Verify the change against the reported duplicate metric and confirm that /metrics returns valid Prometheus-format metrics instead of a 503.