Tart-kubelet restarts no longer orphan running VMs

Tart-kubelet restarts on the Scaleway Apple Silicon CAPI fleet no longer strand long-lived VM workloads. Previously, after a restart the kubelet failed to rebind VMs that were still running, looped every ~30s trying to re-run them, hit “VM is already running” and left the Pod stuck at NotReady, which could block helm --wait --atomic deploys on the xcresult-processor. Running VMs are now rebound at startup and, as a safety net, the next reconcile self-heals any that were missed, so deploys stop hanging after operator rolls.