WiredTiger currently clears the metadata operation queue (conn->disaggregated_storage.shared_metadata_qh) during step-down in __disagg_step_down. At the very least, we need to remove that clean up.

We should also investigate whether there is anything else that we need to do to step down properly. This would most likely involve us (1) dropping local versions of newly created stable tables that were not included in any checkpoints, and (2) clearing the stable_value fields in the corresponding elements metadata operations queue, as those tables would need to be re-created if the node steps up again. We would, however, need to coordinate this with the broader step-down effort, as this change may contradict the design (or at least what the design was at the time this ticket was created).