Shadows use scene near/far

[lilleyse]

For #2594

• Added a few more demo options
  • terrainCast - toggle terrain casting shadows (off by default now)
  • fitNearFar - use the near and far computed in createPotentiallyVisibleSet to limit the shadow map bounds. This helps a lot with views that face the ground, and helps a little with horizon views because it uses a tighter near plane. There is still the problem where moving the camera causes large terrain tiles to load, which throws off the near and far computed from the draw commands. I've tried a few different approaches to recognize when this happens, but none are ideal yet.
• Added DrawCommand.receiveShadows. I removed the code that was using this, so nothing is using this right now, but it might be useful to keep. Since DrawCommand.castShadows exists, it doesn't stand out.
• Only draw debug frustum outlines if in freeze-frame mode, or the shadow map is fixed. Removes some visual clutter this way.

[pjcozzi]

Added DrawCommand.receiveShadows. I removed the code that was using this, so nothing is using this right now, but it might be useful to keep.

Perhaps put a TODO with it. We'll keep it when shadows is merged to master if we are using it or it proves useful for debugging.

[pjcozzi]

terrainCast - toggle terrain casting shadows (off by default now)

I'm sure it is outside the scope of this PR, but there are a lot of artifacts for terrain. For a good place to test, use the gecoder to zoom to "Seneca rocks"

Also note that the skirts still cause artifacts:

We could turn off skirts for the depth pass. It would also be an interesting optimization in general to render skirts after the surface geometry to take advantage of early z. Likewise, we could also not draw skirts if we know all the neighbor tiles are the same LOD so it is watertight. Food for thought, we don't need to do all of this for shadows.

[pjcozzi]

There is still the problem where moving the camera causes large terrain tiles to load, which throws off the near and far computed from the draw commands. I've tried a few different approaches to recognize when this happens, but none are ideal yet.

Maybe just use the BV of the tiles that are expected to be loaded, not the low LOD parent tiles. Perhaps try to hack this in first to see how well it works then we can figure out an elegant approach since 3D Tiles would also need changes.

[pjcozzi]

Just those comments.

[lilleyse]

Updated. The new approach for finding the shadow near/far is working better.

[pjcozzi]

Have you tried this on Mac? Here's the default view (note the black shadow map) on my new AMD MacBook Pro:

[lilleyse]

I've only been testing on Windows, I'll try this out on OSX tomorrow.

[pjcozzi]

Also, it's not specific to this branch. Here is the shadows branch:

[lilleyse]

I'm getting a framebuffer incomplete error in Chrome. When I fallback to the color texture, it works fine. Firefox is working, so maybe this is just a Chrome issue.

[lilleyse]

Added a fix to #3711.

[lilleyse]

Cherry-picked into #3711