Dataset module for v2

[camillebrianceau]

In this PR, I refactored the dataset module with the following changes:

• I kept the API/dataset_test.py in the PR to provide an example of how the dataset is used.
• There is now a single CapsDataset for all extraction modes.
  Future implementations in ClinicaDL will introduce:
  • ConcatDataset
  • PairedDataset
  • UnPairedDataset
    These will inherit from CapsDataset.
• Preprocessing is now represented as an object (using pydantic). This object describes:
  • The preprocessing steps performed using Clinica.
  • The image type being used.
• Readers are the objects responsible for reading data from CAPS or BIDS. These will eventually be replaced by ClinicaIO once its development is complete.

The extraction process has been moved to the Transforms module. Each extraction type (ROI, Image, Slice, and Patch) is now represented as an object and behaves like a transform.

I added unit tests. However, tests for Transforms and Preprocessing are not included, as these components are expected to change in the coming weeks.

[thibaultdvx]

The extraction process has been moved to the Transforms module. Each extraction type (ROI, Image, Slice, and Mask) is now represented as an object and behaves like a transform.

You meant Patch, didn't you?

[thibaultdvx]

Great work, thanks @camillebrianceau!!

[thibaultdvx]

I'm curious, why did you feel the need to create this reader? Is it temporary?

Because I thought that we would handle multi-cohort with multiple CapsDataset, concatenated with ConcatDataset. And it seems to me that each CapsDataset could have its own CapsReader.

Tell me if I'm not clear enough!

[camillebrianceau]

Yes I didn't explain but it was just a first draft. Then we put the CapsReader in the CapsDataset so it is not useful anymore but since we’re not entirely sure about our change, I prefer not to delete it

[thibaultdvx]

Ok, LGTM!