[libc][math] Optimize C23 half precision pow function

`powf16` implementation currently performs all computations in double precision. To improve performance it should use float whenever possible, finding an alternative way to compute for integer `y` would also be ideal

https://github.com/llvm/llvm-project/pull/159906#pullrequestreview-3814710951

Start by locating the libc math implementation of powf16 and review the linked pull request discussion for context. Compare where the implementation currently uses double precision, then determine which computations can use float and how integer y should be handled. Done means the optimization is implemented without changing powf16’s required behavior.

Written by the indexing model from the issue text.