Cesium z-up extension

[ggetz]

Fixes #1324

@lilleyse Please let me know if you have any initial feedback.

[lexaknyazev]

Couple optional comments.

• This seems to be the first extension without an extension object. Maybe, it's worth to mention in the readme that it doesn't exist just in case.

• an implementation must treat the mesh geometry as z-up

  I assume that everything else (node and IBM matrices) should be treated as usual, right? Maybe, the readme could also emphasize that z-up applies to all mesh geometry.

[emackey]

I have to ask, is an extension really needed here? Could one just place a parent node with a 90-deg rotation at the top of the node tree, rather than applying this extension? Are there performance implications for adding such a parent node?

What's really the difference between a model that specifies this extension, vs a model that has a root node with a rotation added to it?

[lilleyse]

Conceptually there isn't any difference, but the purpose of this PR is to avoid that extra node.

For someone who has WGS84 referenced data in their glTF it is a lot easier to tell them to include this extension than describe the changes they need to make to the node hierarchy.

It also allows glTF and 3D Tiles to share the same coordinate system. Although this may not be important broadly, it is really convenient in 3D Tiles pipelines.

[emackey]

/cc CesiumGS/3d-tiles#307

[ggetz]

Closing in favor of solution in CesiumGS/3d-tiles#307