Patches attached.
I also tried to sort equal length codes in such a way that codes that
start and end with 1 bits are paired with values that occur less
frequently as suggested by Michael in
https://ffmpeg.org/pipermail/ffmpeg-devel/2025-April/341917.html. Yet
strangely this worsened compression (for the majority of the fate tests
affected by it). The patch can be found here:
https://github.com/mkver/FFmpeg/commits/mjpegenc_huffman/

- Andreas