The resharding coordinator uses OSI replay protection for initialize commands but does not perform a noop write on step-up, so participants may still accept in-flight initialize commands from an old primary.

When a coordinator steps up and recovers a state beyond kInitializing, it should bump and persist the OSI, then send a noop retryable write to all participants(also known as causality barrier) and wait for majority. This ensures participants reject any stale initialize/abort commands from a rogue primary.