Reorganize how normalized instance attributes are handled

[ptrgags]

Fixes #10766

As of the latest main, instance feature IDs weren't working because they were incorrectly being dequantized in GltVertexBufferLoader.

As described in #10766 (comment), we thought it would be better to not dequantize in the loader at all, only dequantize when absolutely necessary. It turns out there's only one such case where we need to dequantize on the CPU: ROTATION attributes with normalize: true. Translations and scales are not quantized according to the table in the EXT_mesh_gpu_instancing spec.

To summarize, this PR:

• Removes CPU dequantization code from GltfVertexBufferLoader and GltfLoader
• Only dequantizes rotations in InstancingPipelineStage

EDIT: Whoops forgot to add Sandcastle links:

• Sandcastle from the issue -- it should look like a gradient in this branch:
  
• Sandcastle to try the test model -- NOTE: Since instancing bounding spheres use the translation min/max and don't take into account the rotations (that would be more expensive), the top-most box sticks out a little too much and gets clipped with some camera views.

[cesium-concierge]

Thanks for the pull request @ptrgags!

• ✔️ Signed CLA found.
• ❔ CHANGES.md was not updated.
  • If this change updates the public API in any way, please add a bullet point to CHANGES.md.

Reviewers, don't forget to make sure that:

• Cesium Viewer works.
• Works in 2D/CV.

[ptrgags]

Okay, this is now ready for review. @sanjeetsuhag could you review when you get a chance?

[sanjeetsuhag]

Everything looks good. Seems to be the correct fix. Thanks @ptrgags!