[datasets] made sure that id labels are native python int, since otherwise it could cause problem in the concatenation of datasets