Update gltf loader

[takahirox]

This PR updates GLTFLoader.
With this change, it can load the simple glTF 1.1 models listed here

https://github.com/javagl/gltfTutorialModels/blob/master/README.md

Please refer to the commit logs for the change.
And I added some notes where perhaps I need to explain a bit.

I was focusing on loading public sample models.
There're lot of works left to perfectly follow the glTF 1.0/1.1 specifications.

[takahirox]

For skeletal animation, generally bone should have non-empty name (and unique name in a skeleton).
So I set jointName to bone.name if name in json is undefined.

[takahirox]

If matrixAutoUpdate is false,
translation/rotation/scale of this node won't be affected to world
when scene.updateMatrixWorld(true) is called.
So I think here should be true(default).

[cx20]

I confirmed that the glTF 1.1 tutorial model is displayed in the local environment.
I think that there is no problem, but there is one point to worry a little.
The white cube seems to be displayed in VC.gltf of glTF 1.0.

Unfortunately, I do not know whether it is a model problem or a converter problem or an implementation problem.
I do not understand, but the following may be helpful.
KhronosGroup/glTF#576

[takahirox]

I've confirmed the white boxes by applying this change.

I agreed with this comment

KhronosGroup/glTF#576 (comment)

that blend functions (or fragment shader) of input file is wrong.

Other references
KhronosGroup/glTF-Sample-Models#8
KhronosGroup/COLLADA2GLTF#7

Existing GLTFLoader doesn't support blend functions input and always uses NormallBlending.
So that transparency(blending) looked working correctly even with wrong
blend functions input.

[donmccurdy]

This looks like the cause of #8983. 🙃

[donmccurdy]

This will, I hope, be unnecessary when #10366 is fixed.

NVM, I'm too optimistic there.

[takahirox]

As probably you've noticed, here is not for #10366.

The aim is to give the information to AnimationMixer to find the target node from a scene even if node.name is empty (node.name is optional in glTF).

To guarantee to pass the unique specification , always passing uuid could be safer because glTF allows two or more nodes have the same name.

Alternative is to return the mappings of animation root node(=target node) and animation from GLTFLoader, like this

{
    targetNode1: animation1,
    targetNode2: animation2
}

[donmccurdy]

Yes, I misread. My concern with UUID (mentioned in #10366) is that cloned meshes will be incompatible with the original AnimationClip objects, and there isn't (and really shouldn't be) a mechanism for cloning an AnimationClip. The same concern applies to a direct mapping, because you still can't clone the object easily.

But let's keep what you have here, and deal with #10366 later as you suggest. I think we can find a good solution later.

[takahirox]

Ah, that makes sense. Cloning makes the situation complex...
Yeah, let's discuss later.

[mrdoob]

Thanks thanks!