I'm guessing it's simply because the need for f32::max_value() isn't really great enough to warrant allocating resources to it. I doubt there's a technical reason why the method wouldn't exist.
I have a sneaking suspicion that num_trats::Bounded probably used to be part of std back in the pre-1.0 days, and that at the time it was removed/replaced with inherent methods, it was probably only implemented by integers and not floats.
(or rather, I know it used to be part of std, and I have the sneaking suspicion that it was not implemented by floats!)