MeshUtils.merge( object1, object2 [ geometry | mesh ] ): Merges object2 into object1. Implemented in materials_cubemap_sky.

CanvasRenderer code clean up

Cleaned up the escher cubemap code a bit. Also, I think it looks much nicer without illumination... ?
At first I though the cube around the sphere was x-inverted (unless the in tention is to simulate refraction...) I've tried changing that but the effect is still weird. I guess those textures work mainly for the sphere and not for the panorama.

I reused the code and did another example using the skymap. More to come ;)

I can see some lines at the edges of the planes though. I guess that's antialiasing kicking in. Should be fixed with a proper cube mesh I guess.

CanvasRenderer has a property called `autoClear` which is true by default. By making it false you have control of when the renderer cleans the screen (by calling renderer.clear()). I wonder if something like that can be added to the WebGLRenderer? That way we can render the cube outside the spheres from a diferent camera (centered in the middle of the screen) than the spheres floating.

Also, added `shading` parameter to `MeshBasicMaterial` (environment mapping was being rendered faceted when using `MeshBasicMaterial` because `shading` was missing (and thus Faceted)).

TODO:

  - maybe image array loader should go into TextureCube.js so that people could just reuse the code

  - it may be worth to be able to specify different combination of environment map and underlying material (in single render pass),
    currently color/color map/environment map are multiplied, though addition also produces very interesting looking materials

Cloned materials.html -> materials_gl.html to have easier test for WebGLRenderer.

TODO:

Investigate why this test is so slow. It shouldn't be, materials_shaders.html is more complex and yet it is much faster (36 vs 60 fps).

MeshFaceMaterial with every face different is pretty bad use case for WebGLRenderer, but this still doesn't fully explain slowdown :(.

GouraudShading and PhongShading ⟶ SmoothShading

Render first objects / material groups that use normal blending, then render ones which use additive blending.

Proper solution should be: sort triangles by blending, render all the normal ones, sort additive ones by depth and then render them back to front.

http://www.opengl.org/resources/faq/technical/transparency.htm
http://nehe.gamedev.net/data/lessons/lesson.asp?lesson=08

While this is acceptable in C++ / OpenGL, with JavaScript / WebGL this would kill the performance (you would need to sort triangles in JS and also create new buffers every frame).

Also tried to come up with subtractive / darken blending, but didn't find any satisfactory formula :(.

* Changed THREE.Vertex.positionScreen from a Vector3 to a Vector4 to support clipping.

This is actually tricky due to multimaterials and vertex normals.

The practical effects of "shading" parameter in WebGLRenderer are currently following:

- if your model has vertex normals, now you can "dumb down" its shading by specifying THREE.FlatShading for MeshLambertMaterial and MeshPhongMaterial
- if your model has only face normals, it will always have just flat shading, no matter which shading you set, no escape from this
- if your model has multiple materials, smooth shading always "wins", meaning if there is at least one smooth shaded material in multimaterial group, all other materials in this group will be also smooth shaded (this is solvable by having multiple normal buffers, but seems wasteful, at least for the moment, till there are no practical use cases)

Currently implemented shading / material combinations:

  - MeshNormalMaterial: FlatShading [default], GouradShading
  - MeshDepthMaterial:  no shading parameter in material, uses own specific shading
  - MeshBasicMaterial: no shading parameter in material, uses own specific shading
  - MeshLambertMaterial: FlatShading, GouraudShading [default]
  - MeshPhongMaterial: FlatShading, PhongShading [default]

Removed Loader.js from compiled lib
Fixed examples

CanvasRenderer/SVGRenderer: More code clean up and optimisations.

Changed the DepthMaterial to use this approach instead.

SubstractiveBlending ⟶ SubtractiveBlending

Implemented material blendings (THREE.NormalBlending, THREE.AdditiveBlending, THREE.SubtractiveBlending) in CanvasRenderer
Added MeshNormalMaterial (per face)

Cleaned up CanvasRenderer and SVGRenderer code a bit.

It's commented out, as it breaks shader validation in current Chrome :(.

There seems to be some bug in Chrome / ANGLE making it barf when using "gl_FragCoord" in fragment shader.

Curiously, while it breaks both 9.0.576.0 dev channel and latest continuous build 9.0.583.0, it does work fine in canary build with the same version as dev channel (9.0.576.0).

Go figure. Also in Firefox 4 beta 7 it works fine.

It seems there were some recent changes in handling of "gl_FragCoord" on Windows which probably broke something when fixing other bug:

http://code.google.com/p/chromium/issues/detail?id=59762
http://code.google.com/p/angleproject/issues/detail?id=71

TODO: make isolated test case and submit bug report (if this doesn't go away in next Chrome update).

Not all features from the new material format are implemented yet, though already now it's more powerful than the old format (you can now use  texture also in Basic and Phong materials).

Also started to retire old model format: removed old OBJ converter and meshes that it generated.

TODO:
 - take into account shading and blending parameters
 - transplant cube map support from experimental material

Optimised Particles projection in CanvasRenderer.

Most of the examples working.

Almost done porting CanvasRenderer to the new material system (plus some performance gains on the way :D)
`Color` now handles hex with no alpha byte (probably there is a nicer way of doing it...)