tanh 3x faster, and <1.5 ULP vs 2.0 ULP for master

[oscardssmith]

The speedup comes from using exp instead of expm1 for most x. For small x (less than log(2)/2 or .5 depending on the datatype), just use a minimax polynomial. This solution is faster, more accurate, and simpler. I believe accuracy is strictly below 1.5 ULP for all inputs, but I wouldn't be shocked if a 1.6 ULP slipped in somewhere. Either way, current behavior has over 2 ulps for 0.233002233002233 and 1.95 for 0.46070245f0, so this is a major accuracy improvement in addition to the speed improvement.

[stevengj]

(It looks like we already have test coverage for this.)

[ViralBShah]

Good to also have @simonbyrne take a quick look if possible.

[oscardssmith]

One specific question about this. At this point, I'm pretty sure we aren't using ldexp_exp for anything anymore. I also think that _ldexp_exp could be much more efficiently implemented by putting it in exp.jl. Is there anything to act on wrt that here or should I open a separate PR?

[ViralBShah]

If not related to this PR, then make it a separate PR so that this one can go through.