Split out of WT-18644.

The shared metadata queue panics report only the entry that failed the check. The checkpoint-time violations are detected long after the publish and drop calls that caused them, so the table's sequence of operations has to be reconstructed by hand (see WT-18230, BF-45739).

Change

• Before each queue panic, log at error level under schema_epoch and disaggregated_storage:
  • the queue entries for the table: operation, schema epoch or unpublished, and the deferred, in_step_down_window, before_stable_epoch and stable value flags, capped at 16 entries;
  • one line of state: queue length and unpublished count, stable, last checkpoint and step-down schema epochs, and the leader flag.
• Sites in conn_layered.c: the checkpoint-epoch-refers-to-a-recreated-table panic, the parked creates panic, and the two publish panics (create before the stable epoch, published entry at a future epoch).
• No config values, so the output stays small and below mongod's 10 KB attribute limit.