Proper handling of denormals

[eyalroz]

Denormal, or subnormal, floating-point numbers are those with an exponent of zero. Their implicit leading mantissa bit is 0, so one needs to look at their mantissa to determine their "true" exponent. And since printf()-like functions are required by the standard to print non-zero de-normals with a non-zero integral part before the decimal point - they must determine the true exponent.

This library doesn't do so. The worst case of this behavior is with 0.0 and -0.0: That's issue #54 . This is about properly handling all denormals (and covering their handling by some testcases).