During checkpoint prepare, __checkpoint_disagg_maybe_publish gets the btree's schema epoch associated with its create from the metadata queue. We need to ensure that it works correctly if there are multiple create entries, especially if the table was created, dropped, and then recreated.

We could also bypass the whole process (and make it faster!) by adding a new field to WT_BTREE that tracks the schema epoch of its create, which we can populate from WT_SESSION::publish.