Investigate anat_to_template transform direction in resolve_functional

[nx10]

Summary

resolve_functional in src/rbc/bids/functional.py:90-98 queries for from-MNI152NLin6Asym_to-T1w (the inverse composite warp) and assigns it to anat_to_template. This variable is then passed to resample_bold_to_template where it's used in an ANTs transform chain that expects the forward direction (T1w -> template).

Evidence

Registration (src/rbc/core/anatomical/registration.py):

• forward_xfm: composite created with reference_image=template, maps T1w -> template. Exported as from-T1w_to-MNI152NLin6Asym.
• inverse_xfm: composite created with reference_image=T1w, maps template -> T1w. Exported as from-MNI152NLin6Asym_to-T1w.

Functional resampling (src/rbc/core/functional/resampling.py:167):

base_transforms = [anat_to_template, bold2anat_itk]
# Comment on line 172: "Order (last applied first): motion -> bold2anat -> anat2template"

ANTs applies transforms in reverse list order. With reference_image=template, the last-applied transform (anat_to_template, first in list) should map T1w -> template, i.e. the forward warp.

But resolve_functional queries from-MNI_to-T1w (the inverse warp) for this parameter.

The same issue affects _warp_mask_to_template (line 327), which also receives this value.

Action items

1. Run resample_bold_to_template with both forward_xfm and inverse_xfm on a test case and compare outputs to confirm which direction is correct
2. Fix resolve_functional if the query is wrong (swap to from-T1w_to-MNI152NLin6Asym)
3. Consider renaming forward_xfm/inverse_xfm in AnatomicalOutputs to something unambiguous (e.g. t1w_to_template_xfm / template_to_t1w_xfm) so the direction is always clear from the name
4. Audit other resolve functions (resolve_longitudinal_anat, resolve_longitudinal_func) for the same pattern

Context

Found while investigating #287. The naming inconsistency makes it hard to verify correctness of in-memory handoffs between pipeline stages.