Add space subdivision to enhance ray tracing

[alex-eg]

For now, algorithm uses naïve approach and traverses all objects in the scene to find if ray intersects any of them. It's not very productive, and is a quite critical bottleneck.

All object should be stored in a tree-like structure, which allows us to significanlty reduce number of intersection tests.

There are quite a number of available algorithms:

• Various kinds of BVH
• Octrees
• k-d trees

And so on. I think we should stick to BVH at first, and later may as well try other algorithms, it would be interesting to compare them. But there would be separate issues.