Reword docs for `f32` and `f64`

Better explain the reasoning for the `next_up`/`next_down` integer
implementation, as requested by Ralf.