Meshopt compression alters bounds of skinned meshes in some cases

[hybridherbst]

Describe the bug
It seems that in some cases gltf-transform meshopt alters bounds of skinned meshes and results in files with either incorrect scale or incorrect bounds (haven't been able to verify which it actually is so far).

To Reproduce
Steps to reproduce the behavior:

1. Download and unpack BentCylinder-Meshopt.zip
2. Drop "BentCylinder.glb" into https://gltf-viewer.donmccurdy.com/
3. Now drop "BentCylinder.meshopt.glb" into it
4. Note that the second file displays very differently, the only difference is that it was processed with gltf-transform meshopt.

Without meshopt

With meshopt

Expected behavior
Files display the same.
This might also be an issue with bounds calculations inside modelviewer/gltf-viewer instead of gltf-transform (there's known issues with skinned meshes in model-viewer bounds calculations at least, see here: google/model-viewer#3189), but since in this case the only difference is meshopt compression might be a separate problem (and/or the model scale is incorrectly calculated in model-viewer).

[donmccurdy]

I think it's the issue you mention in model-viewer, as well as donmccurdy/three-gltf-viewer#147 - quantization (which can be applied independently of meshopt) applies a scale to skin IBMs to cancel out changes to the vertex values. This keeps the same scene bounding box, but causes both viewers to compute the bounding box incorrectly.

[hybridherbst]

Understood - added some more info (and an implementation by Mugen where I'm not sure if it ever made its way into the threejs core) to the model-viewer issue.