92 unite registration steps into one transformation

[MarcelRosier]

unites coreg and atlas reg transformations to reduce resampling and provides class to perform inverse mappings from atlas to native space. As of now, only Ants and NiftyReg are supported.
Hence, this resolves #92, resolves #100, resolves #122

[Copilot]

Pull Request Overview

This PR unites coregistration and atlas registration transformations into a single workflow to minimize resampling and introduces a class for performing inverse mappings from atlas to native space (currently for ANTs and NiftyReg). Key changes include:

• New test cases for composing and inverting affine transforms.
• Updates to dependency versions and parameter handling for registration modules.
• Addition of the BackToNativeSpace class and modifications in modality processing to capture transformation matrices.

Reviewed Changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_registrators.py	Added tests for affine transform inversion and composition functions.
tests/test_back_to_native.py	Added tests to verify inverse transformation calls with correct parameters.
pyproject.toml	Updated dependency constraints for antspyx, brainles_hd_bet, and scipy.
brainles_preprocessing/registration/registrator.py	Introduced an interpolator parameter and abstract inverse_transform method.
brainles_preprocessing/registration/niftyreg/*	Modified matrix path handling, affine composition, and inversion functions.
brainles_preprocessing/registration/greedy/greedy.py	Added inverse_transform stub to conform with Registrator interface.
brainles_preprocessing/registration/elastix/elastix.py	Added an inverse_transform method that raises NotImplementedError.
brainles_preprocessing/registration/ANTs/ANTs.py	Updated transform and inverse_transform to support list-based matrix paths and parameter passing.
brainles_preprocessing/preprocessor.py	Added a parameter and logic for saving transformation matrices.
brainles_preprocessing/modality.py	Updated transformation path management and registration matrix naming conventions.
brainles_preprocessing/back_to_native.py	Implemented BackToNativeSpace class to apply inverse transformations.

Comments suppressed due to low confidence (2)

brainles_preprocessing/registration/ANTs/ANTs.py:257

• The 'whichtoinvert' parameter is passed to the transform method in inverse_transform but is not explicitly documented or handled in transform; consider updating the API documentation or implementation to clarify its purpose.

            whichtoinvert=[True] * len(matrix_path),  # Invert all matrices

brainles_preprocessing/modality.py:256

• Changing the registration matrix file name to use a 'M_' prefix without a clear extension may lead to inconsistency; consider appending a proper file extension (e.g. .mat or .txt) to match downstream expectations.

        registered_matrix = registration_dir / f"M_{moving_image_name}"

[Copilot]

The _find_transformation_matrix method returns the first match when multiple files exist; consider refining the selection logic or documenting the assumptions to ensure that the correct file is chosen.