powr() implemented inefficiently for non-metal targets

[cheneym2]

It seems like powr() is implemented inefficiently on non-metal targets. If I understand correctly, powr() is only defined for x>=0 but for some reason we coerce x to be abs(x) before delegating to regular pow() for other targets, instead of letting it be undefined.

/// @category math
__generic<T : __BuiltinFloatingPointType>
[__readNone]
[require(cpp_cuda_glsl_hlsl_metal_spirv_wgsl, sm_4_0_version)]
T powr(T x, T y)
{
    __target_switch
    {
    case metal: __intrinsic_asm "powr";
    default:
        return pow(abs(x), y);
    }
}

[jkwak-work]

I think we should simply remove the function powr.
When I added it, I misunderstood what needed to be done and I added bunch of Metal only functions in hlsl.meta.slang.
I wasn't meant to add new intrinsics only because Metal has those.