Kura deploys roll over without write stalls

Bare-metal Kura regions now run with a warm standby and gapless failover: when the primary is rolled, the controller promotes a Ready, runtime-confirmed standby instead of waiting on the old age gate. Promoting a standby that is already an async outbox replica cuts the window of potential misses on the newest artifacts (which regenerate) without stalling writes. The peer plane is also routed directly pod-to-pod across regions and via SNI-passthrough for self-hosted nodes, with the provider keeping a per-region failover IP pinned to a healthy box so the mesh survives a box loss or a rolling deploy.