f32 and f64 support

[tnballo]

Would be nice to add a small_float macro, rounding out (pun intended) numeric primitive support in this library. Unfortunately, determining when a lossless cast from an f64 constant to an f32 equivalent can be done is non-trivial. Due to the specifics of IEEE 754-2008, it’s not as simple as checking (($val as f32) as f64) == ($val as f64).

Maybe understanding how std::num::flt2dec works in supporting Debug for floats might provide some insight. If precision is not explicitly specified when printing, flt2dec helps determine the smallest significant precision (not reflected with an == comparison). But I’m not sure this methodology is 1) actually the desired solution here or 2) possible to implement in a macro_rules macro.