Currently the debug implementation of NotNan is:

        Struct {
            f1: NotNan(
                18.25409836065574,
            ),
            f2: NotNan(
                0.11689901383196721,
            ),
        },

This seems unnecessarily verbose: when debugging structs that contain NotNan floats, we are probably interested by the values, without being particularly interested by being reminded at every step that they can't be NaN (I mean, we see that they are not NaN in their values anyway).

I would much prefer to read:

        Struct {
            f1: 18.25409836065574,
            f2: 0.11689901383196721,
        },

=> It would be useful to have the Debug implementation directly delegate to the inner float, instead of just being a #[derive(Debug)].

Is that something that sounds ok? (I can PR this.)