Faster Printing for Known Normal and Subnormal IEEE754 Floating Point Values

In many cases it is known that a value is not a special value (NaN, Infinity) so faster printing can be obtained by skipping the special value checks.

I suggest adding the functions `formatNormalFloat` and `formatNormalDouble` to `Data.ByteString.Builder.RealFloat`.

Let me know if this is something you would consider. If so, I will create a PR.