[Aggregated Feature] Segmentation downstream evaluation tasks

[ioangatop]

Core:

• Add Linear and Conv2D decoder for segmentation tasks #167
• Add timm encoder and model wrapper networks #381
• Add SemanticSegmentationModule module #382
• Add TotalSegmentator2D segmentation downstream task #383
• Add offline total segmentator configuration #436

Optional:

• Create a callback to visualise the segmentation results #385
• Update TotalSegmentator2D to fetch all the slides #415
• Add dice score in TotalSegmentator2D task #422
• Allow to use subclasses in TotalSegmentator2D #434
• Add Mask2Former decoder  #438
• Allow to train with int16 data  #442
• Support for multi-level embeddings training in segmentation tasks #500

Run with:

eva fit --config configs/vision/dino_vit/online/total_segmentator_2d.yaml