On a disaggregated follower, after a range truncate of [a, b], re-inserting a key inside [a, b] at a later timestamp leaves the key invisible: a scan or search does not return it.

Expected: the re-inserted key is visible, exactly as on a plain table and on a layered leader. A truncate only removes keys as of its own commit timestamp; a write committed after the truncate must survive it.

A reproducer and a proposed fix are in the comments.