Until now, what is reported is:
- maximum query duration: unfortunately, with enough traffic there will always be outliers which are not representative of the normal user experience
- average query duration: when maximum queries are orders of magnitude larger than the vast majority of queries (ex: hundreds of seconds vs. hundreds of milliseconds), it does not take many outliers to skew the average so much to be non-representative
- score: this is more representative than the average query duration by eliminating the longest ones from the score
Adding a median allows giving a better idea of the query duration distribution. This is at the expense of a larger memory footprint, so make this support optional.