format-stress-test-disagg-switch-data-validation-stepdown-async-1 on amazon2023-stress-tests-arm64 fails with the same signature as WT-18255 although the WT-18255 fix (a3b17ae73e) is an ancestor of the failing commit.

Failing task: https://parsley.corp.mongodb.com/evergreen/wiredtiger_amazon2023_stress_tests_arm64_format_stress_test_disagg_switch_data_validation_stepdown_async_1_f1f6b7e56bb0d730dbdbd9434b84ffa101e8281f_26_10_09_00_33_11/0/task
Commit: f1f6b7e5 (BFG-3710746), host i-070a70ad6670b23ee

Symptoms:

file:WiredTigerSharedHS.wt_stable, eviction-server: __evict_server, 278: Cache stuck for too long, giving up: Connection timed out
cache full: no
cache clean check: no (62.715%)
cache dirty check: no (15.312%)
cache updates check: yes (11.997%)
cache dump: total found: 1944.38 MB vs tracked inuse 1783.89 MB
total dirty bytes: 477.84 MB vs tracked dirty 481.00 MB
total updates bytes: 341.26 MB vs tracked updates 368.56 MB

An earlier error in the same run: __wti_disagg_set_crypt_key, 540: set_key timestamp 1715773 must be strictly greater than the stable timestamp 1715809: Invalid argument

Related: WT-18255 (closed), WT-16295 (open, cache stuck over dirty trigger).

Next steps: confirm via the stack trace and cache dump whether this is the same root cause as WT-18255 or a different path that leaves the updates trigger stuck.