Normalize per-instance event metrics policy

[josecelano]

Goal

Make tracker_usage_statistics control metrics processing for an individual public HTTP or UDP listener, without suppressing objective events or UDP ban enforcement.

Concrete UDP Failure Example

With two UDP listeners, one metrics-disabled and one metrics-enabled, one announce to each currently yields REST API udp4_announces_handled = 2, not 1.

The REST API reads this counter from the UDP server metrics repository. One application-wide UDP server event bus feeds that repository with no per-listener metrics policy. The old producer-side event suppression model is no longer valid because UDP server facts are also consumed by an independent banning listener.

Design

Always emit objective HTTP/UDP events with canonical listener identity. Metrics listeners filter disabled listeners before mutating shared aggregate repositories. Banning remains independent and receives relevant cookie-error events from every listener.

Prerequisites

• Define canonical runtime service identity #2036 canonical service-role and configuration-instance identity.
• Migrate runtime service registry metadata #2041 registry metadata migration exposing canonical identity for started services.

Scope

• Normalize HTTP core, UDP core, and UDP server event publication.
• Filter metrics in listeners, not producers.
• Preserve aggregate REST API and UDP operational metrics.
• Add focused, integration, and progressive manual verification, including duplicate port-zero blocks.

Acceptance criteria

• Metrics-disabled HTTP/UDP listeners emit facts but do not update aggregate metrics.
• Metrics-disabled UDP listeners still contribute relevant facts to shared banning.
• Filtering uses canonical identity and works for repeated 0.0.0.0:0 blocks.
• REST aggregate and UDP operational metrics remain available.
• Relevant tests, linter all, and manual evidence pass.

Related to #2035, #2036, #2041, and #1419.