ENH: Fix buffer ownership and a redundant copy in SimpleITK interop

[blowekamp]

Follow-up to #6811. Fixes buffer-ownership in image_from_simpleitk and removes a redundant copy in simpleitk_from_image.

image_from_simpleitk wrapped a numpy buffer via image_view_from_array, giving the itk::Image a pixel container that does not own its memory, which is unsafe for Graft() and in-place filters. It now reads pixels with sitk.GetArrayViewFromImage (no copy) and builds the ITK image with itk.image_from_array (deep copy into an ITK-owned buffer) — same one copy overall, with correct ownership.

simpleitk_from_image read the ITK image with itk.array_from_image (a deep copy) purely to hand it to sitk.GetImageFromArray, which deep-copies again into the buffer the new sitk.Image owns. It now reads with itk.array_view_from_image instead, dropping the redundant first copy.

Why the ownership fix matters

image_view_from_array marks the itk::Image pixel container as not managing its own memory, since it imports a pointer into a numpy-owned buffer. Filters that graft their output onto a pre-existing image, or that reuse/reallocate a buffer in place, assume the container they write into owns its memory. image_from_array deep-copies into a container ITK allocates and owns, removing that hazard.

Testing

black --target-version py310 and the project pre-commit hooks pass on the changed file. Not exercised against real SimpleITK locally (no SimpleITK/itk build environment available). The in-tree simpleitk_protocol.py test does not cover either function directly; per #6811 it stubs only the geometry accessors, by design (round-trip fidelity is owned by an out-of-tree suite).

[dzenanz]

Looks good. For the other conversion direction, it seems that buffer is copied twice. Maybe replace array = itk.array_from_image(image) by array = itk.array_view_from_image(image)?

[hjmjohnson]

@greptile review

[greptile-apps]

Greptile Summary

The conversion updates preserve image pixels after the source image is modified or released, while avoiding an unnecessary intermediate copy.

Confidence Score: 5/5

No blocking failure remains.

The exercised SimpleITK and ITK conversion flows retained independent pixel data after source mutation and deletion.

T-Rex Logs

What T-Rex did

• Created an isolated runtime with ITK 5.4.7 and SimpleITK 2.5.6 and loaded the staged PR extras.py, which uses compatibility aliases for ITK type names.
• Ran the ownership conversion probe script to exercise bidirectional image conversions, mutate sources, delete them, and force garbage collection.
• Validated pixel preservation: both conversion directions preserved the original pixels, as shown in the baseline and post-mutation captures.
• Noted compatibility alias behavior as part of loading extras.py, bridging ITK 5.4.7 naming with the current PR's type names.

_{Ran code and verified through T-Rex}

_{Reviews (2): Last reviewed commit: "STYLE: Condense a two-line comment to on..." | Re-trigger Greptile}