Smtv2 model parser: keep float size

I keep the size of the float type when I parse a float constant so that I can rely facts and not on luck when I search for the initial float ty.