Occlusion issue with changes to log depth equation

[TJKoury]

Issue:

When adding primitives that are far (at least 11000000 m in one axis) away from ellipsoid center, occlusion is broken.

This commit seems to have changed the behavior.

Sandcastle example:
Example

Browser: FF, Chrome

Operating System: Linux

[lilleyse]

It seems like the bug only happens when the scene is rendering two frustums. @kring any ideas?

[kring]

So the immediate problem is that, in the far frustum, the globe, depth plane, and centers of the points are rendered with a range (czm_currentFrustum value in the shader) of: [90000000, 99999998430674944]. But the point outlines are rendered with a range of [100000000, 99999998430674944]. Wat?! Investigating...

[kring]

Hmm ok so it's the opaque frustum near offset. For log depth, it adjusts the near plane by 10%, or in this example 10 million meters! So basically the depth values are just rubbish. When log depth is off, we only adjust the near plane by 0.01% instead of 10%, and if I change the code to do the same for log depth this bug goes away. @lilleyse (or anyone else) do you have any idea why we need such a huge near offset with log depth? Here's the line I'm referring to:

cesium/Source/Scene/Scene.js

Line 1611 in 740ef62

return this._frameState.useLogDepth ? 0.9 : 0.9999;

[lilleyse]

@kring not sure, this was the last conversation on github about it #5851 (comment). I kinda took that change for granted.

If 0.9999 works we should probably just switch.