rf: Split fieldmap registration into its own fit step

[effigies]

The goal of this PR was initially to get SDC reports added to the ASL section of the reports when reusing minimal derivatives. The main thing that needed to happen was to reconstruct the fieldmap in the aslref space, which is part of the unwarp_wf workflow, deep inside the coregistration reference logic.

This PR applies nipreps/fmriprep@32478f8 (part of nipreps/fmriprep#3423) to split up the unwarp_wf into three nodes, and pulls the ReconstructFieldmap node into the new stage 3.

This needs applying to fMRIPrep, as well.

• rf: Separate fieldmap registration from coreg ref generation nipreps/fmriprep#3467

[effigies]

A couple notes for reviewers.

[effigies]

This is needed to relax the requirements for a transform list in ResampleSeries, so it can just SD unwarp a single volume.

[effigies]

The ~= reflects that we try quite hard to maintain backwards compatibility during patch releases.

[Copilot]

Pull Request Overview

This PR refactors the ASL fit workflow to isolate fieldmap reconstruction into its own stage and updates key dependency versions.

• Updated version constraints for fmriprep, sdcflows, and smriprep in pyproject.toml
• Split unwarp_wf logic into distinct stages: added Stage 3 for fieldmap reconstruction, Stage 4 for coregistration reference, and updated Stage 5 logging
• Removed the old init_unwarp_wf import and reorganized workflow connections in fit.py

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
pyproject.toml	Bumped fmriprep, sdcflows, and smriprep to use ~= version syntax
aslprep/workflows/asl/fit.py	Split ASL fit into separate unwarp, fieldmap, and coreg stages; updated node imports and connect calls

Comments suppressed due to low confidence (3)

aslprep/workflows/asl/fit.py:486

• The ReconstructFieldmap node is used here but not imported; please add the appropriate import (e.g., from sdcflows.workflows.apply.fieldmap import ReconstructFieldmap) at the top of the file.

aslref_fmap = pe.Node(ReconstructFieldmap(inverse=[True]), name='aslref_fmap', mem_gb=1)

aslprep/workflows/asl/fit.py:553

• [nitpick] The Stage 4 block duplicates much of the workflow connection logic used elsewhere; consider extracting shared setup steps into a helper function to reduce repetition and improve readability.

if not coreg_aslref:

aslprep/workflows/asl/fit.py:656

• [nitpick] The log message for Stage 5 is ambiguous and similar to Stage 4; consider clarifying it (e.g., 'Stage 5: Adding ASL-to-anatomical registration workflow') to distinguish between stages.

config.loggers.workflow.info('Stage 5: Adding coregistration workflow')

[tsalo]

This LGTM. Just want @mattcieslak to take a look as well before merging.