perf(weather): apply the infrared cloud transfer from a lookup table, in place

The Satellite clouds transfer ran a power curve per pixel and wrote the
result back from a full-size copy. On the 4096x2048 images the 3D Tiles
host draws, that held the page for about 150 ms per image. The scale
depends only on the brightest channel, so it is now worked out once per
byte value and applied to the canvas pixels in place. Output bytes are
unchanged.

Co-Authored-By: Claude Fable 5.2 <[email protected]>

src/layers/weather/infraredImage.test.mjs

@@ -26,6 +26,28 @@ function canvas() {
   return output;
 }
 
+test('the transfer is applied to the canvas pixels in place, with no copy written back', () => {
+  let read = null;
+  const output = {};
+  output.getContext = () => ({
+    drawImage() {},
+    getImageData: () => {
+      read = { data: new Uint8ClampedArray([194, 0, 0, 200]) };
+      // A second full-size buffer copied over the first is what this guards.
+      read.data.set = () => {
+        throw new Error('the canvas pixels were overwritten from a copy');
+      };
+      return read;
+    },
+    putImageData: (pixels) => {
+      output.pixels = pixels;
+    },
+  });
+  processInfraredImage({ width: 1, height: 1 }, 'filtered', () => output);
+  assert.equal(output.pixels, read);
+  assert.deepEqual([...read.data], [194, 0, 0, 98]);
+});
+
 test('decoded pixels are processed once on a same-sized canvas', () => {
   const image = { width: 1, height: 1 };
   const output = processInfraredImage(image, 'filtered', canvas);