Converts floating point coefficent $f$ to rational by the
simple expedient of outputing 10^k*f/10^k for appropriate
$k$. Does no reduction of numbers.  In particular this may cause overflow in
old versions of lrs input (and I'm not sure about cdd).