[shimV2] Fix guest/VM exit on their own

[rawahars]

Summary

• Once a utility VM exits or its in-guest agent dies, container and
  network cleanup has nothing left to act on, yet the old code kept
  returning errors and looping on resources that are already gone.

• Export the closed-bridge sentinel and wrap it around the underlying
  transport failure when the bridge is killed, and add a shared helper
  that classifies the host-side conditions meaning the compute system
  is no longer available for modification. Container resource release
  (combined layers, every category of guest mount and device, and the
  explicit state delete) and both legs of network endpoint and
  namespace removal now treat these two signals as a successful
  release and move on. The VM exit watcher also collapses the bridge
  on termination so in-flight waits unblock instead of hanging until
  context cancellation, and task deletion is allowed once the VM has
  reached the terminated state so containerd can drain stale tasks.

• Along the way, migrate the shim's task service registration to the
  v3 task API, and stop logging the noisy JSON parse error emitted
  when GCS writes a non-JSON trailer (such as a runtime panic dump)
  on shutdown, since the raw payload is already surfaced by the
  terminated-output path.

Testing

• Created custom LinuxBootFiles which introduces a UVM panic after 45 seconds. Ran multiple pods in the same UVM and when the UVM exited automatically, we saw the state being marked as NOT_READY in the containerd output.