TilesRenderer: Traversal overhal plan

[gkjohnson]

Related to #667, #668

Next

• If REPLACE tile's children are not in the frustum then the parent tile should not be in the frustum.
• Make sure behavior of error is correct. Currently the loader will load and display tiles one level over the error target.
• Compare times and optimize update times back to previous release.
• Reduce number of traversal calls if possible
  • Merge some of the functions
  • Push tiles for toggling onto a stack rather than traversing again
• Make sure "ADD" behavior is working correctly
  • See use of "ADD" in markVisibleTiles that assumes the tile with the "ADD" refine is added rather than the children.
  • Consider whether the traversal logic for "allChildrenLoaded" & "recursivelyLoadNextRenderableTiles" need to account for ADD refinement.
  • Check other ADD cases
• See if perf can be improved for "flat" tile sets like the france model
  • Can we avoid calculating error in some cases? Lazy generate?

Goals

• Load OSM tiles from Cesium OSM tile set with "ADD" refinement loads very slowly #663
• Load France tile set from Loading 3D Tiles of France buildings takes long iTowns/itowns#2335

Behavior Changes

• REPLACE tile should load next set of visible children only up to the used set
• SSE should always be calculated for traversed tiles

Plans

• Adjust precomputed tile flags to indicate __hasContent, __hasRenderableContent, and __hasUnrenderableContent.
• Rename traverse methods to be more accurate to their functionality
• Adjust frustum determination function not force immediate children to be used if there are no renderable parents.
• Simplify and fix skipTraversal and markUsedSetLeaves implementation.
• Fix recursivelyLoadTiles function to only load up to what should be visible / used.

[gkjohnson]

Performance changes from 0.3.36 release using the google maps aerial example as a baseline:

Before
~0.85ms
visible tiles: 179
lru cache size: 517

After
~0.91ms
visible tiles: 179
lru cache size: 513

[gkjohnson]

Closing in favor of #678, #679