WT-15575 surprised me that a follower node can issue a checkpoint call. But what a checkpoint on a follower does is (apparently) create a "local" checkpoint which would include any changes to local files, and no shared tables are written. If there are no local files, it's effectively a no-op. It seems kind of weird that we even allow the possibility of local files. If we've checkpointed them, can we recover those files after a failure? And if we can't, what's the point of checkpointing – should it be an error to checkpoint with local files? It's kind of false advertising.

Apparently the server, and our own tests, issue checkpoints on a follower - they've never been "fixed" to not do so, and it would be pretty disruptive to change.

And if that's the case, we should at least mention in doc for checkpoint that on a follower, a checkpoint will not write any layered objects.