2.18.0-beta.23 — a compose update no longer runs out of the time we gave it

One regression from beta.21, reported by @famewolf in #2.

Fixed

• A compose update no longer runs out of time because we gave it more grace. beta.21 passed --timeout <DOCKER_STOP_TIMEOUT> to the compose recreate so a slow container would stop on its own terms instead of being killed after Compose's 10s default. The grace was right; the wait around it stayed a flat 120 seconds, so the time left for pull, create and start fell from 110s to 60s — and a service that has to rejoin a VPN network namespace ran out of it, reporting timed out after 120 seconds.

  Measured against real Docker: a container that ignores SIGTERM spends the full 60s stopping under --timeout 60, and 10s without it. The wait now contains the grace it grants (stop_grace + 120), the same shape the standalone path has always used.

  2.17.1 is unaffected — it never passed --timeout at all. Only beta.21 and beta.22 carry this.