Summary: Return an error instead of aborting when truncate is called while the step-down timestamp is set

Problem

• __wt_layered_truncate has WT_ASSERT_ALWAYS(step_down_timestamp == WT_TS_NONE); it fires in release builds.
• Per-key writes in the same window are routed to ingest or mirrored and straddling transactions get WT_ROLLBACK.
• test/format disables ops.truncate under disagg.stepdown_async to avoid the abort.

Scenario

• Leader in planned step-down; any thread calls WT_SESSION::truncate on a layered: table (MongoDB: pre-image and change-collection removers not verified to stop before the timestamp is reserved).

Definition of Done

• Replace the assertion with an error return (WT_ROLLBACK or ENOTSUP, to be decided) so the caller can retry after the role change.
• Python test: truncate during the step-down window returns the error and leaves no partial state.
• Re-enable ops.truncate under disagg.stepdown_async in test/format or file a follow-up if the drain still cannot account for truncates.