Add more granularity to query/time|size buckets
This is an attempt to increase the precision of the latencies
and bytes reported between 1000 and 10000. From our observations
the values of the metrics, after processed by the histogram_quantile
function, are not inline with what logged by druid.