Make float division generic

Currently there are two separate functions for for `f32` and `f64`.
Combine them and extract some repeated functionality to a trait, which
will make adding `f128` easier.

Author: tgross35