Upgrade t-digest and HdrHistogram library versions

[salvatore-campagna]

Upgrade both t-digest and HdrHistogram libraries to the latest version.

[github-actions]

Documentation preview:

• ✨ Changed pages

[elasticsearchmachine]

Pinging @elastic/es-analytics-geo (Team:Analytics)

[elasticsearchmachine]

Hi @salvatore-campagna, I've created a changelog YAML for you.

[not-napoleon]

We haven't upgraded to t-digest 3.3 because it changed the way it calculates the second quartile. For an even number of data points, it returns the (N/2)th element instead of an average between the (N/2-1)th and (N/2)th elements. This is different from what t-digest 3.2 did, as you can see from the changes you've had to make to the boxplot tests. We've considered this breaking, as it will cause aggregations to return different results than before the change.

[not-napoleon]

I left a note to that effect (although apparently lacking in detail) on the upgrade ticket: #73928

[salvatore-campagna]

@not-napoleon yes I saw that but I am not convinced by the fact that this change is a breaking change. I mean, if the quantile (or percentile) is an estimation of the actual value....why should we consider that a breaking change?

Moreover, why V7 compatibility test check the actual values returned? Should't compatibility be something about endpoints, parameters, formats...instead of the actual values returned? Especially considering that percentiles and quantiles are estimated?

Also, most tests do calculations using a very low number of samples that probably makes results not meaningful. Am I wrong?