WebP lossless encoder: prefix groups, color cache size, lazy matches

The VP8L encoder (lossless WebPs and the alpha of lossy ones) now
- groups the main image's 16x16 blocks by their symbols (k-means by the
  bits their clusters' codes would take, 4-16 groups, from the blocks
  farthest apart) and writes an entropy image with a prefix group each,
  if that is estimated to be smaller;
- picks the color cache's size (0-10 bits) by the estimated bits of the
  literals and hits, all sizes in one pass;
- delays a match by a pixel if the next one starts a longer one;
- tries predicting images of 17-256 colors instead of indexing them, and
  doesn't subtract green from images without red and blue (alpha planes).

Sizes vs cwebp: lossless 113.7% -> 107.2% (5 images), lossy alpha at
effort 2 113% -> 94%, at effort 4 134% -> 111%; every file decodes
exactly in dwebp. Lossless encoding takes about 65% longer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lfv41PqHEGbU7SLg5onpG1

Author: matmen

CHANGELOG.md

@@ -13,6 +13,14 @@
   effort 1 (3.5% at effort 2) and within 1-4% of `cjxl -e 7`'s (up to 9% for screenshots); effort 0 is within 1.5%
   of `cjxl -e 3`. A 1920x1080 photo takes about 1 s at effort 0, 5 s at effort 1, 10 s at effort 2.
 
+### Improvements
+
+- Lossless WebPs (and the alpha of lossy ones) are smaller: the encoder groups blocks of 16 pixels by their
+  statistics (prefix codes per group, like libwebp's entropy image), picks the color cache's size, delays matches by a
+  pixel where the next one is longer, and predicts images of up to 256 colors instead of indexing them where that's
+  smaller (smooth alpha). Lossless files are about 6% smaller (within about 7% of cwebp's), the alpha of lossy ones
+  about 17% (at the default effort smaller than cwebp's). Lossless encoding takes about 65% longer.
+
 ### Fixes
 
 - `AVIFEncodeOptions` is exported (with the new `JXLEncodeOptions`).