While investigating something else, I noticed that calibrated JUNGFRAU data is chunked with a chunk shape of 1 × 1 × 1 × 512. I believe this is a mistake - it means that each single module image is stored in 1024 tiny chunks. This PR makes it 1 × 1 × 512 × 1024, or one image per chunk, like we do for AGIPD.
This makes the files about 5% smaller in my test, and makes writing much quicker (~20 seconds to ~5 seconds per file). Writing took longer than processing before, so this is a significant speedup. Reading will also be quicker, assuming people usually read complete images (it will be slower if you read a single row of pixels).
cc @ahmedk