Update class indices in MoNuSAC to make compatible with generalized dice metric

[nkaenzig]

Monusac evaluation is not compatible with current metrics implementation:

File "eva-worktrees/eva/src/eva/vision/metrics/segmentation/generalized_dice.py", line 56, in update
    target = _utils.index_to_one_hot(target, num_classes=self.orig_num_classes)
  File "eva-worktrees/eva/src/eva/vision/metrics/segmentation/_utils.py", line 48, in index_to_one_hot
    tensor = torch.nn.functional.one_hot(tensor.long(), num_classes=num_classes).movedim(-1, 1)
RuntimeError: Class values must be smaller than num_classes.

The issue was that the ambigous class was mapped to a 255, which is problematic when converting the target tensor to one-hot format.