When using many partitions, this can be a bottleneck (O(n^2) growth with number of partitions) - also it seems DataStoreUtils.constraintsToByteArrayRanges() set maxRanges to 0, and mergeIntersections doesn't use max ranges.