Improve core performance

[mikedilger]

...perhaps by making the core functions branchless.

approx_eq_ulps(), probably the most used function, is pretty short and probably quite easy to make branchless.

approx_cmp() and approx_eq_ratio() are disturbingly long functions IMHO and probably hold promise for plenty of other kinds of optimizations.

Hi,

I want to work on this, Is it ok if I add benchmarks in benches folder ?

[mikedilger]

Go for it.