Hubs no longer get stuck in a failed state when startup races shutdown

When a part of the mesh was starting up at the exact moment its host was being torn down — for example during a pod restart — it could observe the teardown mid-initialization and record itself as permanently failed instead of simply shutting down. Everything served at that address then answered with an error until it was recreated.

That window is now recognized for what it is: a shutdown, not a failure. The affected hub winds down cleanly, error logs no longer report routine teardown as an initialization failure, and a genuine initialization fault still surfaces loudly exactly as before.