Problem

test/format decides whether precise checkpoints are enabled in several places, based on the requested setting, cache size, timestamps, table type, and in-memory configuration. However, the shared DisAgg helper later unconditionally adds precise_checkpoint=true to the connection configuration, bypassing format's disable decisions without updating its internal flag.

As a result, precise_checkpoint=0 does not disable precise checkpoints in the normal DisAgg test path. Format can believe they are disabled while WiredTiger has them enabled, so its initialization and runtime behavior do not match the connection's actual mode.

It is also unclear whether disabling precise checkpoints is valid for DisAgg. The test helper forces them on, but format still has policies that turn them off. We should establish the intended requirement rather than silently maintaining contradictory settings.

Desired outcome

• Clarify whether DisAgg requires precise checkpoints. If required, handle explicit disable requests and incompatible configurations consistently; otherwise, honor explicit disable requests.
• Unify configuration resolution so format's internal flag, connection settings, and runtime behavior agree during creation and reopen.
• Document the precedence of user requests and compatibility policies, and add regression coverage for the agreed behavior.

Failure that exposed the issue

format-failure-configs-test failed on Amazon2023 ARM64 and Ubuntu2004 at b013d1d844 because precise-checkpoint timestamp initialization was skipped while the connection had precise checkpoints enabled:

https://spruce.corp.mongodb.com/version/wiredtiger_b013d1d844aae30be24aa1f19cc765d5b57c16dd/tasks

Detailed consequences, reproduction steps, and validation results are provided in a separate comment.