Implement C23 log2p1

[Paul Zimmermann]

       Dear Joseph,

I tried your patch on a x86_64, it applies cleanly, everything compiles
cleanly, and I can use the new log2p1 function in both binary32, binary64,
double-extended and binary128.

As for accuracy, here are the largest errors I found for rounding to nearest:

float 0x1.a69b4ap-2 1.88 ulps
double 0x1.a7b725780ff2cp-2 1.87575 ulps
long double 0x5.dda3a1f4c4020ebp-4l 4.04684 ulps
Float128 -0x7.fff3a57fdd2666bcecba33bd89f4p-4 3.34125 ulps

Probably some values in libm-test-ulps should be updated.

Best regards,
Paul