Improved WebGLRenderer reflection/refraction/fresnel examples. Now that the renderer is faster we can aswell move 'em ;)

In most cases RGB format is enough (RGBA is needed for example for lens flares to work ok with postprocessing).

It should be safe just to set material color object references, no more updating of color string in every frame for every material color.

Problem were attributes which were enabled for sprites and lens flares, but then not used.

I just delayed enabling till they are not rendered for the first time. There is probably some cleaner way how to do this.

Added remembering of Blender exporter settings between sessions. Also fixed Blender exporter crashing on scenes without world.

Settings are remembered in JSON file "threejs_settings_export.js" stored in Blender's temp folder. It should survive restart of Blender.

Stencil and Sprite examples now using RequestAnimationFrame.

Added handling of mesh colliders to scene loader. Changed Blender scene exporter to make mesh colliders invisible.

This makes lens flares work with postprocessing.

Had to change lines color example to use clearAlpha = 1 to get back bloom.

Removed pink square method from lens flares in opengl (no partial screen occlusion)

Problem was having too many concurrent workers opened at once. Garbage collection apparently works very slowly on workers. Solution was to close them manually, after model data is sent to main thread.

Also some progress with Blender scene exporter. Some scenes now export correctly, though monkey test scene got worse, even with less hacky code.

- Fixed TorusKnot face orientation.
- Added UVs to Cylinder (Cylindrical Mapping) 
- Fixed Lathe UVs

There is something wrong with texture mapping, u needs to be inverted all the time. Something to researching on.

Rewrote lens flares to not use readPixels. There are different solutions based on availablity of vertex textures. If available, occlusion calculation is more exact and handles partial occlusion on screen edges. If not available, occlusion calculation is less exact (because it's made on each fragment so it has to be fast) and partial occlusion by screen edge is not possible (same reason).

It's not possible for the lens flare to update on visibility any more, only its screen position. Potentially, we could add offset+factor for opacity/scale/rotation that is dependent on visibility in shader. Or just custom shaders.

Anyway, it's way faster than readPixels :)

Mainly trying to avoid creation of new object per frame and also some variable names clarifications and whitespace cleanups.

Not finished yet.

MarchingCubes.generateGeometry got forgotten when refactoring vertex normals.

Also some whitespace and titles cleanup.