The model-based verifier should be able to inject crashes during specific parts of a checkpoint to ensure that we are properly testing the corresponding failure recovery.

This would most likely have a form of adding a fail-point into the checkpoint code itself, which would be enabled via debug session configuration. To make the location of the crash flexible and repeatable, perhaps the fail-point parameter could be a numerical parameter, which specifies roughly how far along the checkpoint process should be for the checkpoint to crash. Maybe the new fail-point could replace debug.checkpoint_fail_before_turtle_update.

test/model can probably then take advantage of this feature via a new workload operation, or by modifying the checkpoint operation.