WebGPURenderer: Fix furnace test energy loss for intermediate metalness values

[mrdoob]

Related issue: #32190

Description

Fixes energy conservation issues in the physically-based lighting model when metalness is between 0 and 1.

The furnace test was losing energy for intermediate metalness values due to incorrect application of the Fdez-Agüera multiscattering formula, which is designed for pure dielectric OR conductor F0 values, not blended ones.

This fix follows the same approach used in the WebGLRenderer and matches the implementation used in EEVEE/Frostbite:

1. Store base diffuse color (before metalness reduction) and compute diffuse contribution separately
2. Compute specular color for dielectric path (F0 based on IOR) separately from the blended specular color
3. Compute multiscattering separately for dielectric (F0=specularColor) and metallic (F0=baseColor) paths
4. Mix the results based on metalness
5. Use the dielectric path for diffuse energy conservation

Changes:

• Added diffuseContribution and specularColorBlended PropertyNodes
• Updated MeshStandardNodeMaterial and MeshPhysicalNodeMaterial to compute material properties correctly
• Modified PhysicalLightingModel.computeMultiscattering() to accept F0 parameter
• Updated PhysicalLightingModel.indirectSpecular() to compute separate multiscattering paths
• Updated all BRDF calls to use appropriate color values

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	356.24 86.47	356.24 86.47	+0 B +0 B
WebGPU	610.29 171.21	610.61 171.3	+320 B +94 B
WebGPU Nodes	608.9 170.95	609.22 171.04	+320 B +96 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

	Before	After	Diff
WebGL	487.87 121.22	487.87 121.22	+0 B +0 B
WebGPU	679.12 186.42	679.39 186.53	+273 B +113 B
WebGPU Nodes	621.11 169.71	621.11 169.71	+0 B +0 B

[mrdoob]

@sunag looks good?

[mrdoob]

Oh, looks like webgpu_loader_gltf_iridescence broke.
I'll continue tomorrow...

[sunag]

@sunag looks good?

Yes, that's great.