Add compute functions to ExponentialHistogramState

[JonasKunz]

Part of #135625 , follow up of #136075.

Makes ExponentialHistogramState mimic the TDigestState provided functionality used by aggregations, so that we can build the drop-in replacement HistogramState consisting of both. This will then allow us to apply e.g. percentile aggregation on exponential histograms in addition to T-Digests and a mix of both.

We also had to implement a centroids() functionality, which is implemented by returning the mean values of the populated histogram buckets. Based on my research, centroids() is only used in the boxplot aggregation in order to define the length of the whiskers, where this usage should be fine.

[elasticsearchmachine]

Pinging @elastic/es-analytical-engine (Team:Analytics)

[felixbarny]

would it make sense to pre-allocate the list using centroidCount?

[felixbarny]

I'm a bit confused by this.
This had me believe that you start with the lowest values and go up to the highest ones:

elasticsearch/libs/exponential-histogram/src/main/java/org/elasticsearch/exponentialhistogram/FixedCapacityExponentialHistogram.java

Lines 42 to 46 in a0f415d

	// They store all buckets for the negative range first, with the bucket indices in ascending order,
	// followed by all buckets for the positive range, also with their indices in ascending order.
	// This means we store the buckets ordered by their boundaries in ascending order (from -INF to +INF).
	private final long[] bucketIndices;
	private final long[] bucketCounts;

I guess the last sentence in the comment is wrong then? The indices are ascending but the highest index for the negative scale has the lowest value. Did I get that right?