Bidirectional cluster drift repair — A new "Pull from peer" button in the Cluster UI allows a node to request its peer to push all config tables to it. This fixes ghost-delete drift scenarios where a row was silently removed from one node and the standard repair (push-local-to-peer) would have propagated the deletion. The peer's authoritative copy is now restored without manual SQL.
RepairDrift full-sync safety gate — The orphan-row deletion phase of RepairDrift now requires an explicit ?mode=full_sync parameter. The default repair mode only performs safe upserts and never deletes rows on the peer.