Remove most redundant WebGL calls

[stalgiag]

resolves #3794

This adds some caching to avoid redundancy with many of the most common GL state updates. This puts a little more stress on the CPU but on three different devices (2 laptops, 1 mobile) this PR doubles the framerate with the manual test examples. The number of GL calls is typically reduced by 100-200%.

GL call statisics (Before and After this PR):
performance/spheres -- before: 20,0004 -- after: 10,002
performance/triangles -- before: 17,004 -- after: 9,002
geometryLit -- before: 4,801 -- after: 1,644
geometryUnlit -- before: 1,257 -- after: 462
textures/multipleTextures -- before: 214 -- after: 73
textures/transparentTexture -- before: 101 -- after: 36
geometryImmediate -- before: 373 -- after: 148

Redundancy is removed with calls to:

• gl.uniformXX()
• gl.clearColor()
• gl.enableVertexAttribArray()
• gl.enable()/gl.disable()
• gl.depthMask()
• gl.blendEquation()/gl.blendFunc()

Two improvements that would help further but aren't present in this PR:

• The uniforms have to be tested against the cache. This isn't terrible for simple data structures but for the matrix arrays this can be sub-optimal on the cpu side. It would be better if each uniform was a data structure that tracked its modified status.
• I spent some time trying to implement VAOs (via extension) in the handling of attribute BufferDefs. This proved too difficult to pick apart and would require a refactor. The end result would be more readable code with less redundant calls to gl.bindBuffer() and gl.vertexAttribPoint() but is beyond scope for this.

Small unrelated changes made while working on this:

• Cleanup old manual test examples that are no longer needed
• Add comment to explain repetition between RendererGL.GL and RendererGL.drawingContext (consider future removal of RendererGL.GL)

[aferriss]

@stalgiag thanks for cleaning up the old examples! This looks great, I'm excited to try it out. Did you happen to record any fps while you were testing?

[stalgiag]

@aferriss I haven't recorded any fps. In my testing, I usually saw a 50% - 150% increase in FPS depending on the sketch and device. I have only tested with my phone and laptop in Chrome but I will do more extensive testing before merging and post the results here.