Fix anisotropic ibl bend factor

[issam3105]

Objective

Fix incorrect anisotropic IBL normal bending for KHR_materials_anisotropy.

The previous formula computed 2^(2^x) instead of the x^4 bend factor described by the glTF anisotropy reference implementation. This caused anisotropy to affect areas where the anisotropy texture strength should be zero, visibly producing square texture bounds in assets such as Khronos AnisotropyDiscTest.

Solution

Replace the incorrect nested pow expression with the explicit fourth-power bend factor used by the Khronos glTF Sample Viewer

Testing

Reviewers can visually test this with the Khronos AnisotropyDiscTest and CompareAnisotropy.

Showcase

Before this PR

After

Ref Khronos

Before

After

Ref Khronos
(In this asset, KHR_texture_transform is applied to the normal map, so the normal detail may not line up exactly with the Khronos viewer even after the anisotropy bend factor is fixed)

[olukowski]

Seems the original code assumed pow(a, b) computes b^a, which isn't the case. I was briefly thinking about whether something like:

let bendFactor = 1.0 - anisotropy * (1.0 - roughness);
bent_normal = normalize(mix(bent_normal, N, pow(bendFactor, 4.0)));

would make more sense, but then I realized plain multiplication is probably faster.

So looks good!