SH-2240 WIP on removing lots of string comparisons that were added to deal with exploding amounts of non-built-in GL state

SH-2488 Fix for shiny and alpha not interacting well on rigged attachments (and fix for assertion when rendering model import preview)

- don't use F64 except where needed (should really factor out calls to gluProject)
- get rid of sorting by texture in favor of sorting by matrix (no sort needed, geometry is already matrix sorted as a result of frustum cull tree traversal order)
- unroll matrix sync inner loop and cache MVP and normal matrices

SH-2240 Better fix for beacon rendering -- let LLRender take care of optimization around joining chunks of line segments together into one draw call

SH-2240 Fix for crash when rendering beacons and Debug GL enabled -- flush every 128 beacons to keep from hitting the end of the immediate mode wrapper buffer.

SH-2454 Fix for attachments casting shadows/showing up in water when "show me in mouselook" is disabled

SH-2031 Fix for stall in image update -- don't use the CPU to convert a height map into a normal map -- use the GPU instead via a shader.  Also, WTF glFinish?

SH-2481 Potential fix for viewer detecting incorrect amount of vram available when using certain video cards.

SH-2428 Make GCC 4.4 happy by using LLVector4a striders with LLAlphaObject subclasses -- also vectorize particle VBO updates

SH-2447, SH-2525, SH-2276 Strip out defunct global illumination code, add accounting for how much memory is taken up by LLRenderTarget, fix crash on login in bindGLIndices (work around driver bug that doesn't respect VAO state WRT to index buffers), remove some unused render targets, remove some unused shaders, make it possible to run a fullscreen session

SH-2276 Fix for failed assertions reported from debug-enabled viewers.  Also, remove "Basic Shaders" as a user settable option (shaders are always on if GL version >= 2.0, always off otherwise)

SH-2244 Don't use multisample texture (ever), and don't use GL_DEPTH_COMPONENT32 (use GL_DEPTH_COMPONENT24)