Motivation

test_layered_checkpoint08 contains two unbounded while True poll loops. When the engine fails to make progress (see WT-17700, where the dhandle sweep server stalled for 60+ minutes in the palite/disaggregated config), the test spins silently until a task-level timeout fires — the 2h Evergreen idle timeout, or the 1800s per-test timeout in the parallel_checkpoint hook task (WT-17687). This wastes CI time and produces a misleading "task-timed-out" signature instead of a clear test failure.

Scope

This ticket is the test-robustness follow-up only. The substantive engine bug — why the sweep server stalled in the palite config — is tracked separately under WT-17700 (Foundations).

Approach

Bound both poll loops with a 60-second deadline and a descriptive assertion:
* The sweep-wait loop (dh_sweep_dead_close). With close_scan_interval=1 and close_idle_time=1 a dead handle is closed within a couple of sweep cycles (~1-2s), so 60s is generous even under ASAN/loaded CI while giving ~120x faster feedback than the 2h timeout.
* The wait_for_checkpoint_start() loop (checkpoint_state). The checkpoint thread is already running and the checkpoint takes >=10s due to timing_stress_for_test=[checkpoint_slow], so the state should flip within a fraction of a second.

Definition of done
* Both loops in test_layered_checkpoint08 bounded with a clear assertion on timeout.
* Test still passes on the happy path.