Smarter job division

[Erdk]

Right now each go-routine takes strip of pixels to compute and works only on it, if it finishes its work then exits. The idea is to divide viewport into fixed chunks (64x64 or 128x128) and thread pool of go-routines would work on it.

[Erdk]

I'll be a big one, first:

• prepare architecture doc: client-server
• one server - many clients
• server only distributes jobs and track progress
• clients register workers (cpu, maybe gpu later)
• server: grpc to communicate with workers, http to present status (maybe editor later)

[Erdk]

Actually it's bigger, than I thought. For now I'm moving it to the vFUTURE milestone. I think I'd look on this after finishing "Ray Tracing" book series.

[Erdk]

For now instead of simple "divide render viewport into num goroutines" viewport is divided into 16x16 patches and work queue is created from them. Goroutine take job, process it and updates result image. This loop contines until all jobs from work queue are processed.