Lifecycle metrics omit empty template labels, and a crash during a delete reports the delete operation - #2356

- Add reverting to ate.actor.state to match ateapipb.ActorState and RevertActor state-change events.
- Update ate.actor.lifecycle.operation.duration attribute requirements and notes for create and delete, which do not resolve ate.sandbox.class or snapshot attributes and omit template keys on a failed delete.
- Update rpc.* examples, emitters, and code anchors, plus ate.stats.kind=final and ate.actor.operation.name=unknown descriptions to match current ateapi, ateom, and credprovider behavior.

…provider

- crashBoundActor records a lost worker during DELETING as the delete operation, not unknown. A delete with any_state can run on an actor that still has a worker.
- lifecycleOpAttrs omits the template keys when the workflow holds no actor. Before this change, a failure before or in the load step recorded the template keys as the empty string.
- The lifecycle histogram note no longer repeats each attribute condition. The template keys now state that a failed load omits them. The pool keys now state that a delete never records them.
- The rpc.server.call.duration note names the kubernetes-secrets binary as credprovider. blind_spots now lists the GCP Secret Manager plugin, which has no metrics.
- The lifecycle row in docs/observability.md names ate.snapshot.fidelity, which replaced the snapshot scope.

otelgrpc v0.70.0 sets rpc.system.name on rpc.server.call.duration and on rpc.client.call.duration. On the client histogram it also sets server.address and server.port from the remote address of the connection. The registry did not list these keys.

The registry now defines the three keys and the two histograms reference them. server.port is absent on a call over a unix socket. The brief of server.address says that atelet calls each ateom on a socket path that holds the worker pod UID, so on atelet the key has one value for each worker pod.