AGIPD: Fix dtypes and compression in output file

• Use the correct dtypes for trainId, pulseId, cellId & mask data
• Compress gain & mask data with gzip level 1 (as before)
  • gzip compression is fairly slow: this adds about 30% to the correction time for the data I tested
  • I will experiment with lzf compression, which may be faster

This is meant to fix a problem discussed starting here: https://git.xfel.eu/gitlab/detectors/calibration_workshop/issues/122#note_186424