Fix empty traversal for mixed empty and non-empty tiles

[ptrgags]

Fixes #9356

This PR fixes a bug in empty tile traversal code where sometimes a parent tile wouldn't refine if one of its children is a long chain of empty tiles followed by a tile with content. The old logic would assume the parent shouldn't refine because there was a renderable descendant, and refining would create a hole. However, some cases (like in #9356 and #11564) actually require the hole to render properly.

I went through and redid the logic of that function, and was able to simplify and document it so it's easier to follow.

The biggest difference is that now only tiles that would be rendered for the current camera view are examined, before it would traverse deeper in the tree if there are a lot of empty tiles.

@jjhembd could you review?

CC @lilleyse

[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.
• ❔ Unit tests were not updated.
  • Make sure you've updated tests to reflect your changes, added tests for any new code, and ran the code coverage tool.

Reviewers, don't forget to make sure that:

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

[ptrgags]

I'm going through the Sandcastles in the gallery for further testing. I'm noticing some issues:

1. for Google Photorealistic 3D Tiles, when I move the camera, the view gets blurry when waiting for tiles... so something's not quite right
2. For 3D Tiles Next Photogrammetry Classification, it takes quite a bit longer to load and hits the memory limit, that doesn't happen on main

[ggetz]

@ptrgags Is this something you'd expect to follow up on immediately? If not, please close the PR for now. Thanks!

[ptrgags]

I know the approach in the PR as it stands now is not the correct approach, I get tiles popping in and out with Google Photorealistic 3D Tiles.

It might be worth re-trying @lilleyse's original idea of just returning true. I think there might be some subtleties there indicated by unit tests when I tried it the first time.

For now, I'm closing this PR.