Rethink telemetry enrichment for actor workload telemetry

Telemetry emitted by an actor currently inherits the Kubernetes enrichment default: stamp`k8s.pod.*`, `k8s.node.*`, `k8s.deployment.*` describing the worker pod the actor happens to be running in.

That default is a not a good fit for Substrate, and applying it blindly is probably wrong. For an actor, pod and node are not identity, they are placement. In ordinary Kubernetes those coincide, so OpenTelemetry's semantic conventions conflate them. In Substrate they don't, and the conflation produces telemetry that is expensive, unstable across resume, and can be misread by standard tooling.