Frustum experiment

For the log depth / [multi frustum](https://github.com/AnalyticalGraphicsInc/cesium/issues/4381) stuff, here's another idea for frustum/projection. But I guess there are too many things like putting 3dtiles on top of terrain that make this more complicated that what I wrote...

* Use a projection that maps near far to 1-0 so that all the precision is in the back of the frustum.
* Do a z-prepass and find min max of the depth buffer to use for a tight near far. See section 4 of http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.360.3911&rep=rep1&type=pdf
* post process the depth buffer to remap the gl_FragDepth values assuming the tight near far (may have to use a polygon offset in one of the writes to be on the safe side, i.e. push them back a little assuming the next step...)
* Proceed as normal using the depth buffer to get some early z since its a decent/conservative approx of depth (still write actual depth values).

No files or tests are named. Start by reviewing linked issue #4381 and the cited depth-buffer paper, then trace Cesium's frustum, projection, and depth-buffer rendering entry points. The issue does not define a settled implementation or acceptance test, so completion would require agreeing on scope and validating the proposed pipeline.

Written by the indexing model from the issue text.