fix(mnesia): per-slot volumes hint and harden verification lookup.

Avoid shared Mnesia dir across replicas; make get_or_create_token and CT verify_user resilient to brief cross-node lag.

Co-authored-by: Cursor <cursoragent@cursor.com>

docker/docker-compose.swarm.yml

@@ -76,12 +76,17 @@ services:
           - eventhub
           - eventhub-node
     volumes:
+      # Per-task volume when scaling replicas > 1 (never share one Mnesia dir).
       - type: volume
-        source: eventhub-data
+        source: eventhub-data-{{.Task.Slot}}
         target: /app/data
     deploy:
       replicas: 1
       endpoint_mode: dnsrr
+      update_config:
+        parallelism: 1
+        delay: 30s
+        order: stop-first
       restart_policy:
         condition: any
       labels:
@@ -222,8 +227,8 @@ networks:
     driver: overlay
 
 volumes:
-  eventhub-data:
-#    name: 'eventhub-data-{{.Task.Slot}}'
+  # eventhub-data-{{.Task.Slot}} created by Swarm from service mount template
+  # (do not use a single shared volume for multi-replica Mnesia)
   prometheus-data:
   grafana-data:
   traefik-logs: