Implement `power` with float exponents

Currently, `int.power(a, b)` and `float.power(a, b)` use recursive implementations, and thus only support integers for `b`. As an exception, I've explicitly added an algorithm for square root taken from [`nix-math`](https://github.com/xddxdd/nix-math), which applies when `b == 0.5`, so that the `square_root` function still works. Still, would be nice to have a more general algorithm, which can be hard as Nix doesn't have built-in power.