make expected input shape for segmentation masks more forgiving

[leanderweber]

Description of the problem

The accepted input shapes for segmentation masks (s_batch) seem very strict currently. E.g, for data of the shape (32, 3, 224, 224), only segmentation masks of exactly shape (32, 1, 224, 224) are accepted. I believe quantus could handle some additional shape matching to increase usability. I.e., shapes such as (32, 224, 224) or (32, 224, 224, 1) should also be acceptable, as they could easily be converted.

Minimum acceptance criteria

• [ ] allow for more input shape freedom (within reason) for segmentation masks
• [ ] write additional test cases for these shapes
• pls review @annahedstroem