Y-Flip in Image pass causes inconsistencies

[Vipitis]

mentioned in #30 (comment)

The finally y flipping of the image renderpass is done in the fragment stage. However this can lead to inconsistencies with some functions like dFdy.

A simple diagnostic repro has been provided by ttg via Discord:

void mainImage (out vec4 O, vec2 f) {
  vec2 u=f/iResolution.xy;
  float x= u.x<.5 ? f.y : float(int(f.y+u.x*4.)/2);
  int i=int(u.x*4.);
  O.xy=dFdy(vec2(1,-1)*x);
  O.z=float(fract(u.x*8.)<.04);
}

Solutions to explore:

• ignore it (maybe add notice)
• rewrite rules
• flipping the output in CPU
• additional render/compute pass
• find a solution to do it in the gui?