Add lossy conversion for NotNan<f64> to NotNan<f32>

[MultisampledNight]

Currently, it's not cleanly possible to convert a NotNan<f64> to a NotNan<f32> if you're aware of the precision loss and are ready to take it without unwrapping and re-wrapping again.

So this PR addresses this through adding a NotNan::<f64>::into_notnan_f64_lossy which does that automatically.

I'm not sure about the naming of the method. I'd like to shorten it, and also I'm not sure whether that's a case for using as_ naming, as the Rust API guidelines suggest.

[mbrubeck]

Thanks!

For naming, I think a shorter name like as_f32 or to_32_bit would be fine.

[MultisampledNight]

Thank you for the consideration! I changed the name as_f32 accordingly.