As part of WT-6388, we had to add some logic to overwrite commit timestamps (as well as durable timestamps) of updates as they're being written to history store. For a resolved prepared update where this difference is meaningful, this could cause trouble.

We should add a statistic to keep track of how often this happens.