This prevents the `date_histogram` from running out of memory allocating
empty buckets when you set the interval to something tiny like `seconds`
and aggregate over a very wide date range. Without this change we'd
allocate memory very quickly and throw and out of memory error, taking
down the node. With it we instead throw the standard "too many buckets"
error.

Relates to elastic#71758