Martin/nvidia fixes

[sandsmark]

Applies the fixes suggested by kwand et al in #592

Just smoketested it for about a day and seems to work fine, but I haven't done any in-depth testing to test the efficacy of the improvements.

[sandsmark]

CircleCI doesn't allow me to see what happened.

[morgoth6]

Take a look at #597 Currently testing with vsync enabled and without both __GL_MaxFramesAllowed and __GL_YIELD set.

[sandsmark]

not sure if glFlush really matters in a compositor, though it doesn't hurt I guess.

[codecov]

Codecov Report

Merging #595 (4092469) into next (1d6a014) will decrease coverage by 0.05%.
The diff coverage is 14.28%.

@@            Coverage Diff             @@
##             next     #595      +/-   ##
==========================================
- Coverage   40.10%   40.04%   -0.06%     
==========================================
  Files          46       46              
  Lines        9450     9454       +4     
==========================================
- Hits         3790     3786       -4     
- Misses       5660     5668       +8     

Impacted Files	Coverage Δ
src/backend/driver.c	31.81% <0.00%> (-1.52%)	⬇️
src/backend/gl/glx.c	37.25% <25.00%> (-0.04%)	⬇️
src/win.c	68.98% <0.00%> (-0.29%)	⬇️
src/picom.c	68.41% <0.00%> (-0.12%)	⬇️

[kwand]

Deepest apologies, but after doing some profiling with Nsight Systems, I'm no longer sure if this does anything at all. (I haven't profiled the exact changes as proposed here, but I suspect the results might not be quite different from profiling with just __GL_MaxFramesAllowed set).

Please see #592 (comment).

[yshui]

Suppressed by #641