Background

Design: WRITING-39975. Epic: SPM-4886. Related: WT-18043.

An earlier version of blind deletes had no test/format coverage, so the damage was not caught. A blind remove must not change data consistency, so test/format can use it broadly and check that.

Change

• Open cursors with blind_remove=true for removes in test/format runs that use layered tables, in single-node and multi-node disagg modes, on both leader and follower. Choose it randomly per run so both paths are covered, and log the choice.
• Do not set it on non-layered tables, where it fails with ENOTSUP.
• The existing data checks must pass unchanged.

Acceptance criteria

• Layered table runs, including multi-node runs across step-ups, exercise blind removes and pass.
• The option is never enabled for non-layered tables.

Depends on

Mirrored writes ticket and WT-18921.