Timing experiments with incremental backup showed wildly different times to open an incremental backup cursor, from a hundred microseconds to over 10 seconds. The reason is that opening any backup cursor takes both the checkpoint and schema locks.

I think we should explore if it is safe to not take those locks when opening a duplicate backup cursor. I think it is safe because those locks protect from a checkpoint being deleted and files going away so that the primary backup cursor can generate its list of files to backup.

None of this is needed on a duplicate backup cursor because the locks will be held when opening the primary backup cursor and that primary cursor being open will protect the checkpoints and files for its lifetime.