dials.import_xds issues with experiment model creation

[jbeilstenedmands]

Following updating the program, in further testing a couple of issues have become apparent:

1. [Now resolved] When importing XDS data where the image data is from a h5 file, (sometimes/always?) the experiment created is understood by dials/dxtbx to have only one image, as seen in the output from dials.import_xds:

--------------------------------------------------------------------------------
Experiment 0
  format: <class 'dxtbx.format.Format.Format'>
  type: <class 'dxtbx_imageset_ext.ImageSequence'>
  num images: 1
--------------------------------------------------------------------------------

This then leads to a crash when a subsequent program tries to load the xds_models.expt file:

File "/dials/modules/dxtbx/src/dxtbx/model/experiment_list.py", line 492, in decode
        imageset = self._imageset_from_imageset_data(imageset_data, models)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/dials/modules/dxtbx/src/dxtbx/model/experiment_list.py", line 412, in _imageset_from_imageset_data
        imageset.set_scan(scan)
    RuntimeError: dxtbx Internal Error: /dials/modules/dxtbx/src/dxtbx/imageset.h(365): DXTBX_ASSERT(index < scans_.size()) failure.

This is the error reported in #2938 (comment) which I can reproduce on recent data processing at Diamond. I have a potential fix coming in dxtbx for this.

2. (Sometimes?) Issues with the geometry that manifest in dials.two_theta_refine/dials.refine not being able to sensibly refine the crystal model, with very large rmsds. Observed on the beta-lactamase (c2sum) and the semisynthetic_multilattice datasets. If testing xia2.multiplex, you can temporarily get around this by setting unit_cell.refine=None.