bevy_pbr: Normalize world_normal

[greytdepression]

Objective

Fixes #6528

Solution

Commit 681c9c6 introduced a bug for certain GPUs where the foxes in the many_foxes stress test example were not shaded correctly. The cause of this bug (at least on my machine) is apply_normal_mapping() no longer normalizing the world_normal vector.

Note that this solution is not ideal as the comment specifically says:

NOTE: The mikktspace method of normal mapping explicitly requires that the world normal NOT
be re-normalized in the fragment shader. This is primarily to match the way mikktspace
bakes vertex tangents and normal maps so that this is the exact inverse. Blender, Unity,
Unreal Engine, Godot, and more all use the mikktspace method. Do not change this code
unless you really know what you are doing.
http://www.mikktspace.com/

We may want to revert this in the future, but for now it seems to fix this bug.

---

Changelog

Reverted part of 681c9c6

[greytdepression]

@hymm @DasLixou @SecretPocketCat can you test this out on your machines?
@superdump do you know if this change might cause any issues in other places?

[hymm]

The foxes seem to be colored correctly now.

Note: On discord @mockersf mentioned that animated_fox example is colored correctly on main. The difference might be that the animated fox is not scaled, but the ones in many foxes are scaled. So the issue might be that we're not dealing with scaling correctly somewhere and normalizing at this stage will just hide the probelm.

[Elabajaba]

I tested animated_fox with a sub 1.0 scale and it slowly gets more glitched out as you decrease the scale from 1.0.

[superdump]

I think #6543 is perhaps the proper fix... perhaps.

[superdump]

The root cause that I fixed in #6543 was that skinned normals were not being normalised in the vertex stage, which was missed in #5666. Also, it turns out that even normalising all normals in the vertex stage and then having them interpolated to each fragment, there is a discrepancy in the length of the normals so I added re-normalisation at the end of apply_normal_mapping().

I'm grateful for @greytdepression looking into this. At the same time, I think this PR incorrectly re-normalises the normal in the fragment shader before normal mapping is applied, which Mikkelsen expressly said should not be done. Nor does it address the scale problem that some were observing. #6543 addresses these problems so I'd prefer to close this one and go with that one. I hadn't seen that this PR had been opened until late in my debugging/fixing process but I want it to be clear that I do appreciate the effort!

[SecretPocketCat]

@hymm @DasLixou @SecretPocketCat can you test this out on your machines? @superdump do you know if this change might cause any issues in other places?

Looks good on my setup.

[hymm]

Works on mine

[superdump]

Closing as #6543 was merged. Thanks again for the effort!

[greytdepression]

Closing as #6543 was merged. Thanks again for the effort!

No problem, it was fun digging around in the source code a bit! :D
I'm happy you found the actual underlying problem! My PR was more of a quick hack to put a bandaid on it