fix(gateway): complete worker registration cleanup

crates/aether-data/adapters/mysql/migrations/20260731000000_cleanup_duplicate_worker_boot_runs.sql

@@ -0,0 +1,22 @@
+-- Worker supervisors are registered as one logical row per task. Older binaries
+-- included the ephemeral gateway instance in the row id, leaving a permanently
+-- running row after every restart. Remove only those system-generated boot rows;
+-- current workers recreate the stable logical rows after migrations complete.
+-- The metadata predicate also replaces task-only rows written by early builds of
+-- this fix that still claimed an instance owner. Delete children explicitly so
+-- cleanup remains complete after imports performed with FK checks disabled.
+DELETE FROM background_task_events
+WHERE run_id IN (
+    SELECT id
+    FROM background_task_runs
+    WHERE id LIKE 'boot:%'
+      AND owner_instance IS NOT NULL
+      AND created_by = 'system'
+      AND progress_message = 'worker booted'
+);
+
+DELETE FROM background_task_runs
+WHERE id LIKE 'boot:%'
+  AND owner_instance IS NOT NULL
+  AND created_by = 'system'
+  AND progress_message = 'worker booted';