OpenGL: boost shader performance for Lanczos3 optimized

[FernetMenta]

See title. Allows HQ scaling on a 4k display.

Some cleanup in shader area.

[fritsch]

Something went wrong from the version yesterday:

Seems like the offsets are wrong (vertical stripes)

Performance wise: Really great. It basically uses half the sampler load and therefore needs also only half the throughput per second. Tested with gputop.

[fritsch]

@lrusak if you find an GLES < 3.1 alternative for textureGatherOffset you could adapt 1:1 - from my pov this is the source of most of the advancement we get here.

[fritsch]

Update: Dithering is broken - the shader works fine :-)

Here is a summary of the work fernetmenta did and was pleased to help testing:

[fritsch]

Load with new Shaderwork (Upscaling 576i50 to 3840 50p):

No skips, no drops.

[fritsch]

Old implementation:

Server drops and skips. You can see the Sampler Load and the Texels needed is double of the new implementation and therefore unusable on slower systems. This could be the revival of the Baytrail :-) or the the intel HD2000 when OpenGL 4 drivers are available.

Test-Setup: Apollo Lake Broxton.

[fritsch]

The squash commit fixes dithering - thx much.

[FernetMenta]

jenkins build this please

[FernetMenta]

UWP-64 error unrelated