Added encoding option to force 32bit encoding for non integer numbers

[hildjj]

I'm assuming I understand your use case, but can you explain it here so we've got it on the record, please?

[hildjj]

Can you please add yourself to the contributors section in package.json?

[janpoltan]

Added option to force encoding of non integer values to 4 byte float.

[hildjj]

After trying to write tests for this, I'd like to discuss what happens with large floats. They are currently encoded as a 32-bit Infinity (fa 7f 80 00 00), which is different than the 16-bit Infinity that cbor.encode(Infinity) produces (f9 7c 00). I see a couple of options:

• Document the current behavior.
• Error.
• Add support for BigFloat's, which probably breaks your use case and will have interesting interactiosn with the cbor-bigdecimal package, which implements bigfloats.

Of those, document is the easiest of course. Error has precedent, there are several ways to cause an Error to be thrown in Encode - the only niggle is how to detect the issue, but perhaps a couple of checks in _pushFloatBE would work (check the buffer after writing for Infinity and -Infinity). I don't really like the third option.

Regardless, we need to change _pushInfinity, _pushNaN, and _pushIntNum's -0 handling to look at useFloat32bit also.