Fix missing basis orthogonalizations in GLSL backend

[pablode]

I've encountered shading artifacts in normal mapped MaterialX materials, and upon closer inspection I believe that these are caused by the missing basis orthogonalizations @Tellusim brought to attention in PR #1049.

However, instead of using a second normalize(cross(..)) call, I've applied the Gram-Schmidt algorithm.

Here are some comparisons from the soon-to-be-released test suite for my glTF to USD+MaterialX converter:

Name	Old	New	Diff
NormalTangentTest
DamagedHelmet
IridescentDishWithOlives
ClearCoatTest

The changes can be tested using following model:
NormalTangentMirrorTest.zip

It requires a recent build of MaterialXView as it makes use of the support for explicit bitangents introduced in PR #1156.

[jstone-lucasfilm]

Thanks @pablode, this looks like a great improvement to me.

Refining it one step further, perhaps we should wrap the newly-added step in an mx_gram_schmidt_orthogonalize helper function? Its input arguments could be the original X and N vectors, and its return value could be the modified X vector. This would also help to make the code more self-documenting for new readers.

One potential home for this new helper function would be:
https://github.com/AcademySoftwareFoundation/MaterialX/blob/a345d134b865a6cdaf63e10cfc870df49ed841be/libraries/pbrlib/genglsl/lib/mx_microfacet.glsl

[jstone-lucasfilm]

@pablode Let's plan to make that additional refinement in a future pull request, and I believe your existing change is worthwhile as it stands. I'll run some visual tests in a local build to confirm your results, and then I believe we can merge this pull request to main.

[jstone-lucasfilm]

Thanks for the fixes, @pablode!