Report latency distributions for each workload function separately

[pkolaczk]

Now when you run a mixed workload, e.g. reads with writes, latte will measure latency of reads separately from the latency of writes.