NibabelWriter fails on 2D channel-less arrays with no target affine (regression in 1.6.0) #9166

Describe the bug

NibabelWriter.set_metadata() raises TypeError: '>=' not supported between instances of 'NoneType' and 'int' for a 2D array written with channel_dim=None when no original_affine / spatial_shape is given. The same call works for 3D arrays, and it worked for 2D arrays in 1.5.1.

spatial_resample now derives the spatial rank from get_spatial_ndim(img) (the MetaTensor.spatial_ndim path). For this input that gives rank 1, so the spatial_size is None and spatial_rank > 1 auto-size branch is skipped. spatial_size stays None and then reaches fall_back_tuple(..., lambda x: x >= 0).

To Reproduce

import numpy as np, tempfile
from monai.data.image_writer import NibabelWriter

for shape in [(64, 64), (8, 64, 64)]:
    w = NibabelWriter(output_dtype=np.float32)
    w.set_data_array(np.random.rand(*shape).astype(np.float32), channel_dim=None, spatial_ndim=None)
    w.set_metadata({"affine": np.eye(4), "original_affine": None})
    w.write(tempfile.NamedTemporaryFile(suffix=".nii").name)

• MONAI 1.5.1: both shapes OK
• MONAI 1.6.1: (64, 64) fails with the TypeError above; (8, 64, 64) OK

Expected behavior

2D channel-less arrays are written unchanged, as they were in 1.5.1.

Workaround

Passing "spatial_shape": data.shape in the metadata avoids the failing branch. Output is identical when no target affine is given.

Environment

MONAI 1.6.1 (also present on dev), Python 3.12.