Improve MDL thin-film support

[krohmerNV]

• push down the thin film parameters to the nodes that can use them, i.e., dielectric, conductor and generalized schlick

• removed additional inputs, that transport thin film parameters, from the public material interface by making them non-editable

• make sure thin_film will not result in undefined behavior when passing zeros as parameters (0 thickness disables thin_film but 0 ior is invalid)

• Vertical Layering, push base deeper than one level, needs review, contains TODOs for mix, add, and multiply (If not 100% sure this is correct, we can drop that commit and postpone.

[pablode]

Hello! I'm seeing a regression when I use these changes with my MaterialX/MDL-based renderer:

Before	After

I've prepared a glTF+mtlx version of this asset in hope that it can be reproduced with the MDL SDK DXR example:
maneki_gltf.zip

[krohmerNV]

I can reproduce it. And it showed another problem as well:
When reading the the normal texture, the output is color and connecting that to a float3 in MDL requires and explicit type construction (because color could be spectral).
The MaterialX document is not passing the validation either:

[09:07:28] [MDL_D3D12] [WARNING] [MTLX] Validation warnings for 'S:\content\gltf\maneki\maneki.mtlx'
Mismatched types in port connection: <output name="normal_rgb" type="vector3" output="rgb" nodename="normal">

@jstone-lucasfilm can you tell how such UsdUVTexture should work in MaterialX?

But I'm a bit surprised about the missing gold here. I'll investigate and keep you posted.

[pablode]

Ah right, I brought up the color3-vector3 issue in #1038 before.

I actually work around this in my renderer by patching the network, but dumped the source too early.

I believe there were talks about adding a <vector3> specialization of the UsdUVTexture node, which would solve this problem: #1070 (comment)

[krohmerNV]

Looking into the regression and it revealed an MDL bug which will be fixed with MDL 1.8. Meanwhile, I added a workaround to detect the problematic case and handle it correctly.

[krohmerNV]

about the color to float conversion, I left a comment with a suggestion in #1038

[kwokcb]

If we want the vec3 version of USDUVTexture how about we bring it up in the MTLX/USD channel and see what folks have to say.
IMO, it's an implementation variant and not a spec change but would need to discuss as it still produces 2 implementation signature variants for USD to consume.

[krohmerNV]

@jstone-lucasfilm and @kwokcb is there something outstanding for this PR? I believe the type of the usd texture return rgb is a different topic discussed in #1038

[jstone-lucasfilm]

@krohmerNV As long as the original regression noted by @pablode has been addressed, I'm not aware of any outstanding issues, and I'll plan to take a closer look at these changes soon.

[jstone-lucasfilm]

Thanks for this improvement, @krohmerNV, and this looks good to me!