Restart of a disaggregated standby with a large schema spends tens of seconds deleting local ingest files, then recreates them. On the M20 DSC run in sls-heavy-schema-dev-startup-stall (https://mongodb.slack.com/archives/C0C2LN7HQG4/p1791584368473459) the drop was median 26s (worst 55s) and the recreate median 96s (worst 2m 40s). The recreate is WT-17915. The drop exists because each ingest table is in-memory and still gets a descriptor file, which the next startup deletes. Do not create that file. The metadata entry is the object.