Implement cld

[eschnett]

This implements the "cld" function as suggested in #7132.

[StefanKarpinski]

This is a nice thorough pull request and the complexity of the implementation makes me inclined to think that maybe we should include this. In particular, the -fld(-x,y) version doesn't work for small unsigned types. I'm not entirely convinced that all of the bittwiddly complex versions are necessary.

[eschnett]

The "bitwiddly" versions are presumably the optimized implementations for integers when some arguments are known to be unsigned. This mirrors the implementation of fld, which looks equally complex, and has equally many versions. Using -fld(-x,y) doesn't work for integers if x or y have extreme or near-extreme values.

(If we wanted to allow wrong results for extreme values, then there exist much simpler and more efficient implementations for both fld and cld, but that doesn't sound like a good idea.)

[JeffBezanson]

Very nice.