Fix clipping plane issue with child tiles with transforms

[ggetz]

Fixes #6600

Part of the problem may be that only the root's transform is used in the Cesium3DTile/contentVisibility function. I vaguely remember the reasoning behind this came up as a PR comment but I can't find the thread.

This is intentional - The tile intersection calculations were correct (turned on debugBoundingVolume in the below example to illustrate). The clipping planes' modelMatrix is relative to the root tile, so we should use that to determine the planes' location in world space.

The problem was that each tile's model shared a reference to the same clippingPlaneCollection (to save memory), and therefore couldn't account for additional child transforms. Added a private additional matrix to model which each tile sets individually so they can continue to share the reference to the clippingPlaneCollection.

Sandcastle example

[cesium-concierge]

@ggetz, thanks for the pull request! Maintainers, we have a signed CLA from @ggetz, so you can review this at any time.

---

I am a bot who helps you make Cesium awesome! Contributions to my configuration are welcome.

🌍 🌎 🌏

[lilleyse]

It looks like the root tile is no longer getting gradually clipped in the demo.

[lilleyse]

modelViewMatrix already incorporates computedTransform, so this might be double applying it and leading to unexpected results.

[ggetz]

@lilleyse Tweaked to use the proper transforms for i3dm and b3dm.

b3dm needs the inverse of the relative transform from the root tile.
i3dm needs the root tile transform to be applied to the modelInstanceCollection model.

[hpinkos]

@ggetz @lilleyse what's the status here?

[lilleyse]

Still working on a fix for the case here: #6600 (comment)

[pjcozzi]

Related to or replaced by #7034? @lilleyse

[OmarShehata]

I built off the solution here, so I'd say replaced by #7304.

[pjcozzi]

@ggetz please close if you are OK with that.