Python xarray conversion: CI never runs test_xarray.py; follow-ups to #6937 #6950

#6937 fixes image_from_xarray() assigning origin and spacing to the wrong axes
for 4D images, a bug present since at least ITK 5.3. It survived because ITK's CI
never executes test_xarray.py. This issue lists the CI fix first, then the
remaining test and code follow-ups, each with explicit instructions.

1. Make CI run the xarray tests (highest value)

Two independent gates turn the test into a silent pass in every CI build:

• Install xarray in the Python pipelines. Add xarray to the pip line in
  Testing/ContinuousIntegration/AzurePipelinesLinuxPython.yml (line 80),
  AzurePipelinesMacOSPython.yml (line 83), and
  AzurePipelinesWindowsPython.yml (line 62). Today the test hits its
  except ImportError: branch, prints "xarray not imported. Skipping xarray
  conversion tests", and exits 0.
• Report the skip instead of passing. In
  Wrapping/Generators/Python/Tests/CMakeLists.txt (test PythonXarrayTest,
  line 223), add
  set_tests_properties(PythonXarrayTest PROPERTIES SKIP_REGULAR_EXPRESSION "xarray not imported")
  so CDash shows the test as skipped, not passed, wherever xarray is absent.
• Wrap 4D images in at least one Python CI build. The 4D block in
  test_xarray.py (guarded by "(<itkCType unsigned char>, 4)" in itk.Image.GetTypesAsList(), line 136) only runs when 4D images are
  wrapped. Source builds default to ITK_WRAP_IMAGE_DIMS = 2;3
  (Wrapping/WrappingOptions.cmake, line 241). Set
  ITK_WRAP_IMAGE_DIMS:STRING=2;3;4 in one Python dashboard configuration
  (Linux is the cheapest). The PyPI wheels already wrap 4D.
• Verify on the CI log that PythonXarrayTest executes the 4D asserts.

2. Tests to add in Wrapping/Generators/Python/Tests/test_xarray.py

All inside the existing 4D block:

• 4D round trip (fails before BUG: Fix axis order in image_from_xarray origin/spacing #6937, passes after; verified locally on
  the ITK 5.4.7 wheel):
  python rt = itk.image_from_xarray(itk.xarray_from_image(image)) assert np.allclose(rt.GetSpacing(), image.GetSpacing()) assert np.allclose(rt.GetOrigin(), image.GetOrigin())
• 4D input stored in non-canonical order: a DataArray with
  dims=["x", "y", "z", "t"] and distinct per-axis coords must give spacing
  (sx, sy, sz, st) and origin (ox, oy, oz, ot).
• 4D vector image (dims=["t", "z", "y", "x", "c"]), guarded by the
  matching wrapped type, to cover the "c" path.

3. Code follow-ups in Wrapping/Generators/Python/itk/support/xarray.py

• Non-suffix dimension sets raise ValueError. ordered_dims = ("t", "z", "y", "x")[-image_dimension:] (line 115) assumes the present dims
  are a suffix of (t, z, y, x). dims=("t", "y", "x"), ("t", "x"), or
  ("z", "x") fail with ValueError: tuple.index(x): x not in tuple. Either
  support them with
  ordered_dims = tuple(d for d in ("t", "z", "y", "x") if d in data_array.dims)
  plus a test, or raise a clear ValueError that names the supported
  combinations.
• Remove the dead sort. After BUG: Fix axis order in image_from_xarray origin/spacing #6937, image_dims is used only for its
  length. Replace lines 112 to 114 with a direct count, for example
  image_dimension = len({"t", "z", "y", "x"}.intersection(data_array.dims)).
• Remove the duplicate direction assignment in xarray_from_image
  (lines 66 and 85 compute the same value).
• Size-1 axes lose their origin. if coords.shape[0] > 1: (line 133)
  skips both origin and spacing for a single-sample axis, so a single time
  point gets origin 0. Set the origin whenever coords exist and keep the
  spacing default for size-1 axes.