Throughput values are not normalized

[brad-lewis]

On a system running a consistent workload, the throughput (k/s) totals should be of the same order for runs of different lengths. This is not the case because normalization is not being applied to the data and datat aggregations:

estat zio  10
...
                                       iops(/s)  avg latency(us)       stddev(us)  throughput(k/s)
 read, asyncr                             24575              298            85376           866152
 write, asyncw                               28              303           214094             1546
 write, syncw                                11              262          1281182             1164


                                       iops(/s)  throughput(k/s)
 total                                    24649           870060

estat zio  1
...
                                       iops(/s)  avg latency(us)       stddev(us)  throughput(k/s)
 read, asyncr                             27264              261            18828            78362
 write, asyncw                              198             9287         -2192869             1078
 write, syncw                                19              156             2638              116


                                       iops(/s)  throughput(k/s)
 total                                    27869            80314

These is a bug in all versions so far and happens when normalization is explicitly turned on (estat zio -n 1).
Additionally when normalization is turned off (estat zio -N 1) the units on throughput are displayed as (k/s)