Can date_histograms better take advantage of data locality?

[jpountz]

Description

Date histograms are one of Elasticsearch's most used aggregations. Given the way we index into data streams, it's quite likely for documents with similar @timestamps to be clustered together.

Could we take advantage of this to speed up date histograms (or effectively ranges, since date histograms often rewrite to ranges) by first checking whether the current doc falls within the same bucket as the previous doc before doing more expensive operations? For instance in the case when a date histogram rewrites to a range, this could help save the binary search on bucket boundaries.

[elasticsearchmachine]

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

[wchaparro]

Added to: #65019