GLSL Processing Blocks

[dorodnic]

Work in Progress - Do not merge

This pull-request introduces auxiliary realsense2-gl module. It can be used to perform common SDK processing tasks on the GPU in a relatively generic way (vendor neutral, unlike CUDA) via GLSL shaders.
In addition, it serves as a proof-of-concept for future SDK extensions beyond core API.

Benefits

• On most platforms this will greatly improve Viewer responsiveness.
• On some platforms, this can be a useful way to reduce CPU utilization and improve overall performance (Jetson series is a great example, and unlike dedicated CUDA optimizations this does not require any extra steps to build)

Limitations

• Mac support is firmly missing. OpenGL support on the Mac is rough and it's unlikely to improve.
• To handle OpenGL context management in a cross-platform way, this PR relies heavily on (and somewhat coupled with) GLFW.
• At least for now, this set of enhancements is not helping at all on low-end GPUs. Perhaps this can be improved with shader optimization and simplification.

When it can be used?

• In general it is rare to come across a system where one approach is hands-down better than the other (probably Jetson is the most extreme case)
• When the goal of the application is rendering, it makes much more sense to do everything on the GPU
• When chaining several GLSL blocks together some boost can be accomplished by avoiding unnecessary copies between main and GPU memory
• It allows to distribute the workload more evenly, freeing the CPU to do other stuff

How to use it?

• Adaptations required to existing code are fairly minimal. Some explicit initialization is required, but otherwise rs2::pointcloud becomes rs2::gl::pointcloud, rs2::colorizer becomes rs2::gl::colorizer, and so on.
• rs-gl example is included to get started.