Optimize OpenGL depth packet processor image flipping

[SirDifferential]

This pull request optimizes OpenGL depth packet processor by reducing the CPU time of downloadToNewFrame(). In the old system textures arrive from the graphics card upside down and they are flipped on the CPU. This pull request adds some tweaks that skip the flipping on CPU and instead parses the texture coordinates as flipped in the first place.

Very Sleepy profiler output before optimization
Very Sleepy profiler output after optimization

If there are issues with this PR, please let me know.

[christiankerl]

uv_flip missing here?

[SirDifferential]

christiankerl: Yes, I think so. There's also a couple in stage2.fs I did some initial tests and spent a while reading what the Z and X tables do, and it indeed seems like they need to be read using the flipped coordinates as well.

However, I also noticed an odd black region in the image that appears every 30 frames or so. I'd say hold this PR for now until I can investigate more.

[SirDifferential]

Er, I accidentally pushed my Intel media SDK changes in the master branch. I'll see if I can fix this somehow and remove them from this PR.

EDIT: I'll make a new PR after I fix the issues. I moved my commits to a new branch, so this PR automatically closed.

[xlz]

While you're at it, can you make it an option to flip the image so existing users would not get unexpected results? You can leave the option variable internal, or setting it from an environment variable, so we can expose this option later through the API.