Add time-series query rewrite to push $limit into $sort (for sort on the metadata field) so that we don't sort buckets that are not necessary

It is safe to assume that every bucket has at least one measurement. We should rewrite [\{$_unpack}, \{$sort}, \{$limit: x}] pipeline as [\{$sort(with limit)}, \{$_unpack}, \{$limit: x}]