fix: Remove checks for unit zooms and symmetric rotations in template warp (#3376)

A user with a custom template is generating warps that have spacing
`[1.5 1.5 1.5]`:
https://neurostars.org/t/use-custom-dartel-template-in-fmriprep/30670/8

The check for unit spacing was mostly intended to avoid weird behavior
where we might generate a bad affine. The interpretation of isotropic
spacing is straightforward, so I think we can relax this. Creating a
docker image to allow for testing.

Author: effigies

fmriprep/utils/transforms.py

@@ -49,7 +49,8 @@ def load_ants_h5(filename: Path) -> nt.base.TransformBase:
     #   * Always return a nitransforms TransformBase
     #   * Construct warp affine from fixed parameters
     #
-    # This should be upstreamed into nitransforms
+    # This has been upstreamed into nitransforms.
+    # Future versions should switch to using nitransforms directly.
     h = h5py.File(filename)
     xform = ITKCompositeH5.from_h5obj(h)
 
@@ -78,14 +79,6 @@ def load_ants_h5(filename: Path) -> nt.base.TransformBase:
     spacing = transform2['TransformFixedParameters'][6:9]
     direction = transform2['TransformFixedParameters'][9:].reshape((3, 3))
 
-    # We are not yet confident that we handle non-unit spacing
-    # or direction cosine ordering correctly.
-    # If we confirm or fix, we can remove these checks.
-    if not np.allclose(spacing, 1):
-        raise ValueError(f'Unexpected spacing: {spacing}')
-    if not np.allclose(direction, direction.T):
-        raise ValueError(f'Asymmetric direction matrix: {direction}')
-
     # ITK uses LPS affines
     lps_affine = compose_affine(T=origin, R=direction, Z=spacing)
     ras_affine = np.diag([-1, -1, 1, 1]) @ lps_affine