Error-free Training for MedMNIST Datasets

Package the 'Error-free training' method into a standard training utility for medical image classification. If this is truly effective at preventing repeated mistakes, it could be a massive gain for MedMNIST researchers.

Suggested repo: PerfectTrain

"Train your classifier to perfection with zero repeat-error logic."

Estimated effort: 30h