Implement more loss functions

[pesekon2]

It would be nice to compare results also for trainings done using different types of loss functions. It is also being stated that boundary-based losses or compound losses could be valuable for imbalanced data.

• distribution-based losses:
  • cross entropy
• compound losses:
  • DiceFocal
• region-based losses:
  • Dice
  • Tversky
• boundary-based losses:
  • Hausdorff Distance

For more, exempli gratia: https://jeune-research.tistory.com/entry/Loss-Functions-for-Image-Segmentation-Distance-Based-Losses