Implicit tile not refining correctly

[ptrgags]

My team was working with a 3D Tiles 1.1 tileset with implicit tiling. Nothing particularly unusual about this one, it's an octree with a box bounding volume. The GE value at the root tile seems reasonable for most of the tileset except this one.

Unfortunately I can't share the data for this one, only a couple select screenshots to show the issue:

When I zoom in on this building, it's still blurry and the tile is huge. However, if you look to the right side, the neighbor tile (same size as this one) has refined once or twice.

Furthermore, if I fly to another side of the building and zoom in very close, it does finally refine, but zooming in further it goes back to blurry again:

Blurry	Better!	Blurry again??

Also note that this does not happen in Cesium for Unreal, the tile refines just fine

This effect seems to be sensitive to camera angle, the problem shows up more/less depending on which direction the camera is facing.

As for things to look into:

• In the debugger, check the SSE computation to see if anything is unusual. To filter for the tile, I would use the implicit tile ID (TODO: determine which tile it is. I think it's one of the level 1 tiles).
• see if this is related to Possible error in computation of geometric error for implicit tilesets #11174

[ptrgags]

I'm starting to look into this. I started by looking at the tiles that are loaded.

Observations:

• The first tile with the problem is (level; x, y, z) = (1; 0, 0, 0) which fortunately is easy to remember
• in terms of the tile.children array, there are 6 available children out of 8. With +x to the right, +y into the screen, +z up, we have the bottom layer available (2; x, y, 0) plus two from the top layer (the two closest to the camera, (2; 0, 0, 1) and (2; 1, 0, 1)).
• However, only 5 out of the 6 children are loaded! The missing one is (2; 1, 0, 1).
• Strangely, only the bounding volume for the missing child appears!
• Also it looks like (2; 1, 0, 1) has no content nor should any of its descendants (at least from a quick visual inspection
• I just noticed that when I zoom in on the building with the issue, there's a tiiiiiny triangle in floating above the lightning rod just barely in the new tile, (5; 14, 7, 8)
  
• oh wait, found a better view, there's supposed to be a tile underneath it, (5; 14, 7, 7) but doesn't always show up
  

Things to check:

• Double check which content files I have on disk to see if the above makes sense
• Inspect the availability bitstream and see if anything is incorrect (it's still possible that this is a data problem)
• Determine why tile (2; 1, 0, 1) is loaded but seems to have no content or descendants.
• Look through the refinement logic in CesiumJS to see if I'm hitting a corner case. it might have to do with the fact that 5/6 children are loaded (at least nominally, that seems incorrect as it is)

[ptrgags]

I had a chat today with @lilleyse, we determined that this is a known issue, #9356 and confirmed that the fix described here fixes this tileset, is more spec-compliant, and is more consistent with Cesium Native.

I'm closing this issue in favor of #9356, I'll start putting together a PR for it.