Does Kudu colocate data sets with identical keys? If so, are there plans to have Impala take advantage of this?