Adapting CLP to Floating-Point Arithmetic