Second part of splitting up dev/gap. Only the in-memory transform; no
reader changes.
- `GapPixels` and `ModuleGaps` configure 2 pixel gaps at the chip
boundaries inside each Jungfrau or Eiger module and, only when module
gaps are set, gaps between modules. Gap pixels get a fill value or, with
`split_counts`, half of the neighbouring double size pixel.
- `insert_gap_pixels` and `gapped_shape` apply the transform to an image
in memory. Gap positions follow detector coordinates, so a ROI receives
exactly the gaps that fall inside it.
- Python bindings, tests and docs for the above.
Chip and module sizes are fixed at 256 and 1024x512, so there is no
detector type or quad argument: an Eiger quad is 512x512 and its only
chip boundaries are at pixel 256.
Reading raw files with gaps comes in the next PR.
