Spacingd crashes when the data dict has a non-dict key starting with '<key>_' (regression in 1.6.0) #9165

Describe the bug

Since #8269 (in 1.6.0), Spacingd.__call__ updates every dictionary entry whose key starts with f"{key}_", assuming each one is a metadata dict:

meta_keys = [k for k in d.keys() if k is not None and k.startswith(f"{key}_")]
for meta_key in meta_keys:
    if "filename_or_obj" in d[key].meta and is_supported_format(...):
        d[meta_key].update(d[key].meta)

Any other value that happens to share the prefix (for example a string image_path, or an int image_index) makes the transform raise AttributeError: 'str' object has no attribute 'update' for NIfTI inputs.

This breaks MONAI Label in practice. Its inference pipeline always sets data["image_path"], and its radiology app uses Spacingd("image", ...), so NIfTI inference fails on MONAI 1.6.x.

To Reproduce

import numpy as np, tempfile, nibabel as nib
from monai.transforms import Compose, LoadImaged, EnsureChannelFirstd, Spacingd

f = tempfile.NamedTemporaryFile(suffix=".nii.gz").name
nib.save(nib.Nifti1Image(np.random.rand(16, 16, 8).astype(np.float32), np.eye(4)), f)
t = Compose([LoadImaged("image"), EnsureChannelFirstd("image"), Spacingd("image", pixdim=(2.0, 2.0, 2.0))])
t({"image": f, "image_path": f})

• MONAI 1.5.1: OK, output shape (1, 8, 8, 4)
• MONAI 1.6.1 and current dev: RuntimeError: applying transform Spacingd ... caused by AttributeError: 'str' object has no attribute 'update'

Expected behavior

Only metadata dictionaries should be updated. Entries that are not mappings should be left untouched.

Suggested fix

Only update entries that are mappings, e.g. if isinstance(d[meta_key], MutableMapping) and .... I'm happy to open a PR.

Environment

MONAI 1.6.1 (also reproduced on dev), Python 3.12, torch 2.x CPU.