Why don't you generalize to other practical floating point sizes?

[yurivict]

For example, single precision floating point number? Maybe, keep the same rules, just allow the user to set different significand/exponent sizes?

Very nice work, by the way! Thank you!