If the WT_SYNC_CHECKPOINT operation errors out in __wt_sync_file, it will not wait for the pending and running reconciliation work to finish. This results in undefined behavior once the code execution moves past the WT_SYNC_CHECKPOINT cleanup phase.

As a solution, we should call __wt_checkpoint_parallel_finish during the error path before the cleanup happens. We can optionally cancel any pending work items by acquiring the queue lock, draining the work queue, and updating the relevant counters.