Calculate multiple percentiles when running the benchmarks

[passaro]

Tell us more about this new feature.

We currently run throughput and latency benchmarks for 10 iterations and collect the average run time. In order to better compare different runs of the benchmarks, we should also collect other statistics across the 10 iterations, like multiple percentiles and/or the standard deviation.