This comes out of the investigation in WT-17617, where this idea was POC'd. On a follower's first checkpoint pickup, which might get the complete set of layered btrees, we shouldn't need to create all ingest tables at once, it would be much better to create them lazily. On subsequent pickups, the number of btrees to be created should be much smaller and the schema lock is already held, so it's easy enough to create those synchronously to the pickup.