Learning discrete distributions with infinite support