fix: restore column-major loop order in error diffusion

The perf optimisation changed x-outer/y-inner to y-outer/x-inner.
With the Floyd-Steinberg kernel this changes which neighbouring pixels
have already received error when the current pixel is processed.
Row-major scan causes error to accumulate horizontally, producing
large buildup along vertical edges that pushes pixels to wrong
bright palette colours (visible as coloured halos).

Restoring column-major (x-outer, y-inner) matches the original
processing order. Bounds check and Float32Array optimisations kept.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: gyng

src/filters/errorDiffusingFilterFactory.ts

@@ -53,8 +53,8 @@ export const errorDiffusingFilter = (
     // Scratch buffer — avoids per-pixel array allocations in palette calls
     const _pix = new Float32Array(4);
 
-    for (let y = 0; y < H; y += 1) {
-      for (let x = 0; x < W; x += 1) {
+    for (let x = 0; x < W; x += 1) {
+      for (let y = 0; y < H; y += 1) {
         const i = (x + W * y) * 4;
 
         if (useLinear) {