At the moment, when checkpoint reconciles a page it will always leave the in-memory page in place with the original set of updates. It can often mark the page clean.

Those pages with updates that just hang out cause the cache to be in a less optimal state.

We should add code that will attempt to swap the page with updates, with an equivalent disk-image that was just generated by reconciliation. That process would be equivalent to scrub-evicting a page that was dirty. It requires that checkpoint tries to get exclusive access to the page, if it succeeds it will proceed with the scrub eviction.