gpu: Draw proper shadows again

The fix in commit 5e7f227d broke shadows while trying to make them
faster.
So use a better way to make them faster.

With the normalized blur radius, we can now conclude that all the values
too far from p.y will cause the gauss() call to return close to 0, so we
can skip any y value that is too far from p.y.

And that allows us to put an upper limit on the loop iterations.

Tests included

Fixes #6888

gsk/gpu/shaders/gskgpuboxshadow.glsl

@@ -114,13 +114,15 @@ blur_corner (vec2 p,
   p /= _sigma;
   r /= _sigma;
   float result = 0.0;
-  float step = max (1.0, r.y / 8.0);
-  for (float y = 0.5 * step; y <= r.y; y += step)
+  float start = max (p.y - 3.0, 0.0);
+  float end = min (p.y + 3.0, r.y);
+  float step = (end - start) / 7.0;
+  for (float y = start; y <= end; y += step)
     {
       float x = r.x - ellipse_x (r, r.y - y);
       result -= gauss (p.y - y, 1.0) * erf_range (vec2 (- p.x, x - p.x), 1.0);
   }
-  return result;
+  return step * result;
 }
 
 float

testsuite/gsk/compare/outset-shadow-corner-blur-huge-nocairo-nogl.node

@@ -0,0 +1,10 @@
+clip {
+  clip: 1400 1400 20 20;
+  child: color-matrix {
+    matrix: matrix3d(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 255);
+    child: outset-shadow {
+      blur: 1000;
+      outline: 0 0 1000 1000 / 1000 0 1000 0;
+    }
+  }
+}

testsuite/gsk/compare/outset-shadow-corner-blur.node

@@ -0,0 +1,7 @@
+clip {
+  clip: 0 0 15 15;
+  child: outset-shadow {
+    blur: 10;
+    outline: 0 0 100 100 / 100 0 100 0;
+  }
+}

testsuite/gsk/meson.build

@@ -119,6 +119,8 @@ compare_render_tests = [
   'opacity-colormatrix-combinations',
   'opacity-overdraw',
   'opacity-overlapping-children',
+  'outset-shadow-corner-blur',
+  'outset-shadow-corner-blur-huge-nocairo-nogl',
   'outset_shadow_offset_both',
   'outset_shadow_offset_x',
   'outset_shadow_offset_y',