In the usec granularity histograms, there is a huge gap between 1000 and 9999us. We should split it into at least 1000-2499, 2500-4999, and 5000-9999 buckets to provide better resolution for these measurements. Similarly, we should split up the <10ms bucket in the msec granularity histograms, as many of our FS level latencies have lately been in that range.

Do this in the disag. integration branch first, as that's where it would be most useful for our performance investigations.