Blitting code refactor

[boberfly]

Thought I'd log it here as an issue:

The current blitting code has a few issues, notably it looks as though there are too many branch checks at the pixel-level where we could refactor the branches at the whole image/tile level and win back performance, as these checks are mostly for determining conversion to/from types that don't change per-pixel on an image.
https://github.com/tangent-opensource/hdcycles/blob/main/plugin/hdCycles/renderBuffer.cpp#L32

The other issue is that currently I think single channel buffers are not blitting right, this is a screenshot of the depth pass at the screen edge (using BlitTile function):

Cheers