Problem

test/csuite/schema_disagg_abort stress-tests disaggregated schema operations across a leader/follower role switch, then SIGKILLs the child and verifies recovery. Its switch mode performs the leader -> follower transition by closing and reopening the connection (workload.c, comment: "graceful step-down is not yet supported") and quiesces all worker threads at the phase boundary.

WT now supports graceful in-place step-down (reconfigure(disaggregated=(role=follower))), and test/format already exercises it asynchronously (format_disagg.c) while worker threads keep committing: arm a step_down_timestamp cut-over, drain in-flight work past it, pin stable, take a step-down checkpoint, reconfigure to follower - no connection close. schema_disagg_abort does none of this, so the graceful step-down path is never exercised under concurrent create/drop/insert load.

Cover both single-node (one connection switching roles in place) and multi-node (separate leader and follower nodes, real hand-off) in this ticket, as the coordinator effort overlaps.

Definition of done

• schema_disagg_abort switch mode performs leader -> follower via graceful async step-down (reconfigure, no connection close), with schema worker threads live throughout the transition.
• Multi-node variant: leader steps down and a separate follower node picks up the checkpoint and steps up.
• Switch mode runs repeatedly with no drain timeout, no EBUSY or panic during the live transition, child SIGKILLed in phase 2, recovery + verifier pass.
• Non-switch mode unchanged (no regression).