CBF origin vectors off by half a pixel

[phyy-nx]

As discussed in dials/dials#2194 (text modified from this comment):

dxtbx is reading the header wrong for MiniCBF. ImageCIF specifies counting from the center of the pixel, not the corner, as shown in these two references:
http://www.bernstein-plus-sons.com/software/CBF/doc/cif_img_1.8.4.html#_diffrn_detector_element.reference_center_fast
http://www.bernstein-plus-sons.com/software/CBF/doc/cif_img_1.8.4.html#array_structure_list_axis

Compare to the relevant lines from FormatCBFMini, where we aren't subtracting 0.5 from the beam_xy parameter.
https://github.com/cctbx/dxtbx/blob/main/src/dxtbx/format/FormatCBFMini.py#L128
https://github.com/cctbx/dxtbx/blob/main/src/dxtbx/format/FormatCBFMini.py#L170

Likewise, FormatCBFFullPilatus uses imgCIF_H, which calls cbflib's get_pixel_coordinates(0, 0), which according to the docs, returns the center of pixel 0,0. So to go from ImageCIF to dxtbx we need to apply a half pixel correction. So with this diff (and after dials/dials#2194 is merged), the image from dials/dials#2190 displays properly:

diff --git a/src/dxtbx/model/detector.py b/src/dxtbx/model/detector.py
index 399fde70..ceeee2a0 100644
--- a/src/dxtbx/model/detector.py
+++ b/src/dxtbx/model/detector.py
@@ -768,7 +768,7 @@ class DetectorFactory:
             cbf_detector.get_inferred_pixel_size(2),
         )

-        origin = tuple(cbf_detector.get_pixel_coordinates(0, 0))
+        origin = tuple(cbf_detector.get_pixel_coordinates(-0.5, -0.5))
         fast = cbf_detector.get_detector_axis_fast()
         slow = cbf_detector.get_detector_axis_slow()
         size = tuple(reversed(cbf_handle.get_image_size(0)))

Note, this is like the cause of the 0.5 pixel offset problem @nksauter worked through in 2020 (see section 2.3.1).

Also note if we fix all the CBF format classes by subtracting this half pixel offset, we'll need to make a corresponding change to cbf_writer, which writes CBFs following the dxtbx convention instead of the ImageCIF convention. It will need to add a half pixel to the vectors it writes.