feat(docker): a swarm service's metrics stay under one key across its tasks (#6718)

#6718 "A swarm container's metrics, normal and anomalies are keyed by its task
name, so they start over whenever the task moves". The operator: "the metrics
should be keys to the service name not the task name... as that's what's
consistent between updates."

CPU, memory, restart count and health now go under "swarm/<service>", so a
normal, its anomalies and an alert rule carry across a redeploy or a move
between nodes. Network rates stay "<host>/<service>", the task id gone, as they
answer who on a host is sending. Tasks of a multi-task service share the key,
so its level reads as an average across them (of 51 live services, only
authentik-worker and portainer_agent run more than one task).

History under task names moves in a background task, 5000 raw samples every
15 s, not a migration: the raw table holds about 2M such rows and the app
would wait on it at boot. Hourly rows combine where two tasks share an hour,
and rows still to review merge to one per metric.

A service's anomalies no longer reach a host's page, as it has no one host;
get_anomaly reads what happened around one from the service's own events.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

steward/core/alerts.py

@@ -65,9 +65,10 @@ def agent_heartbeat(source_module: str, metric_name: str,
     _HEARTBEAT = Heartbeat(source_module, metric_name, offline_after)
 
 
-def agent_fed(source_module: str, host_of: Callable[[str], str]) -> None:
+def agent_fed(source_module: str, host_of: Callable[[str], str | None]) -> None:
     """A source whose samples a host's agent carries; `host_of(resource)`
-    names the host ("nas:/" → "nas", "nas/web" → "nas")."""
+    names the host ("nas:/" → "nas", "nas/web" → "nas"), or None for a
+    resource no one host carries."""
     _AGENT_FED[source_module] = host_of