Add quantiles to summaries

[cyberdelia]

Calculate quantiles in summaries, this makes Summary actually useful for this client. This reuse the default parameters from the Go client to be somewhat consistent.

[dmagliola]

How does this work with aggregation of Data Stores from different processes?

The library used to have Quantile support, and we removed it when we added support for multi-process Web servers.

My understanding at the time is that quantile are fundamentally non-aggregatable (at least not without storing every observation which is not feasible). It's possible this is because of the quantile gem design, which you're not using, and not a fundamental limitation, but i'd like to understand how this works when the observations from different processes get aggregated.

Explanation on the issue with aggregating quantiles in our proposal

Removal of quantile support, also with explanation.

[cyberdelia]

I failed to do a bit of homework. This doesn’t behave much better than the previous implementation in that matter.

But for use cases where the trade-off of not using a multi-process web server is acceptable, collecting quantiles in summaries is very useful (rather than having them do almost nothing).

I’m happy to add a guard preventing declaring objectives in summaries for multi-process compatible data stores alongside some warning in the documentation. I’d be happy to do that with the previous quantile implementation as well.

[SuperQ]

It's easier, and more accurate, to use a Histogram.