Add fast paths for Math(f).pow

[MaxGraey]

I saw some people prefer use Math.pow(x, 2) instead x * x or x ** -0.5 instead 1 / Math.sqrt(x).

Optimally this cases should replaces during std calls simplify transform in compile time. For example LLVM have SimplifyLibCalls transform for that but our Mid-end IR still in progress. So I introduce temporary solution for that with fast paths inside Math.pow / Mathf.pow which will be removed in future.

Binaryen could sometimes reduce this functions with const second argument via DAE pass (dead argument elimination).

Fast and safe cases:

x ** 0   => 1
x ** 0.5 => sqrt(x) + special handling
x ** 1   => x
x ** 2   => x * x

x ** -0.5 => 1 / sqrt(x) + special handling
x ** -1   => 1 / x

// But we can't provide equivalent fast paths for cases presented below
// x ** -2 => 1 / (x * x) <- not safe. May loose precision
// x ** 3  => x * x * x same. May loose precision

[dcodeIO]

Thanks!