Remove padding value and padding token parameters from training configuration

We should remove padding value and padding token parameters from training configuration since we always use the default values. See https://gitlab.com/teklia/atr/dan/-/blob/14a54602f2ce89b8732da01a018765e459cec541/dan/ocr/document/train.py#L110

changed milestone to %DAN-P4: Improve data loading and preprocessing

added Quick Win label

changed the description

assigned to @manonBlanco

mentioned in merge request !164 (merged)

closed