Using native histograms to keep track of request latencies

[pedro-stanaka]

Summary

Using native histograms to keep track of latency, since they give much more precision on quantile estimates.

Also a couple of changes:

• Using seconds as measurement, following Prometheus best practices
• Changing buckets for the metric to use custom values instead of the default buckets from prometheus.