Similar Items: Optimizer-Model Consistency: Full Finetuning with the Same Optimizer as Pretraining Forgets Less