Effects of separate sleep counters

The merge of WT-4216 has had a mixed effect on the performance plots. It still feels like the right change because the second sleep call from WT-4208 was added without resetting the existing counters. Since the merge of WT-4216, the stress tests have shown a large improvement and consistency of the latency of the stress test, but the mixed shrink and mixed balance runs have shown a performance degradation. I have attached a plot. The only change listed in at build #4886 on that plot is the change for WT-4216.