mkFp16

Create FP16 from the value.

Important: we suppose that value has biased exponent, but FP16 will be created from the unbiased one. So, at first, we'll subtract KFp16Sort.exponentShiftSize from the value's exponent, take required for FP16 bits, and this will be unbiased FP16 exponent. The same is true for other methods but mkFpCustomSize.