Fix due to Bill Noon for problem discovered by Ken Manheimer: packing

0.0 as float or double would yield the representation for 1.0!