Adding decimal "digits" in multiplication resulted in signed overflow and
producing wrong results.

  Fixed by using large enough buffers and intermediary result types :
dec2 (currently longlong) to hold result of adding decimal "digits" 
(currently int32).