Improve Prometheus support

[simplylizz]

Currently on each unique randomly requested URL (incl. non-existent) collector will create 18 metrics (15 buckets + sum + count + total), so it looks like this could easily bloat both: collector memory footprint and Prometheus DB if it's configured to scrape all metrics. It would be nice to have an option to collapse all 404 requests into one or maybe two metrics (count + sum should be enough) without URL and/or method within labels.

Also I don't fully understand what is the difference between count and total metrics. Is one just counter type (i.e. always increasing) while another should be affected by sliding window?

Regarding window: https://prometheus.io/docs/practices/histograms/ - as I can see from the Prometheus docs histogram shouldn't have window on exporter side, should it?

Other possible improvements (tbh, I don't need these, but it's worth to mention):

• Optionally use summary instead of histogram;
• Make buckets configurable;