ci(compose): give the worker 60s to drain on SIGTERM

Without stop_grace_period, docker SIGKILLs containers 10s after
SIGTERM. For the huey worker that means an in-flight mail send,
periodic close_due task, or DB transaction can be cut mid-way — the
mail is half-delivered, the row is mid-commit, or the next worker
will reprocess the same task without idempotency keys (which we now
have for mail).

60s matches gunicorn's --graceful-timeout from docker-entrypoint.sh
and leaves headroom for the longest realistic in-flight task.

Author: mxhash

compose.yaml

@@ -47,6 +47,9 @@ services:
     <<: [*service-base, *app-security]
     container_name: struudel-worker
     command: ["worker"]
+    # Give huey_consumer time to drain in-flight tasks on SIGTERM —
+    # docker's default 10s would SIGKILL mid-mail-send or mid-DB-txn.
+    stop_grace_period: 60s
     # python is already in the image; scan /proc instead of pulling in
     # procps just for `pgrep`. Exits 0 if any process has huey_consumer
     # in its argv, else 1.