add latency target option #110
Description
People often want to know "how many queries per second can my cluster handle while keeping a p99 of 20ms?"
We should keep a running sample of the last 10 pXX latency numbers in a long array of 10 elements and dynamically the rate limit to meet that.
We should be very clear that it might take a while to narrow in on the target since response time varies with data set, compaction activity, GC pauses, etc.