Implement film grain synthesis

In theory we could try somehow merging apply_to_row_y and apply_to_row_uv. They are separate because I wrote the simpler y function first. It's not impossible to reconcile their differences, but I'm not sure whether it would be worth the effort. (It would most likely also be slower, since the luma-specific code affects the most pixels and keeping it with as few branches as possible should therefore help)

TODO:

• fuzz
• test on other files than the ones I've thrown at it
• benchmark vs libaom
• parallelize
• SIMD
• test bitexactness (md5) against libaom
• avoid memcpy() by using per-plane refs (requires picture changes)
• add GLSL shader example to docs?