Impossibility of Characterizing Distribution Learning -- a simple solution to a long-standing problem