Add option to save optimizer and scheduler state during training, and to resume training from these states

[alanakbik]

This PR adds functionality to save and load optimizer and scheduler states. You can enable this by setting save_optimizer_state=True in any of the train methods of ModelTrainer (train(), fine_tune(), or train_custom()). When enabled, both the optimizer state and the scheduler state are saved in the model file.

If you load a model that contains these states and pass it to the ModelTrainer, it will automatically load them if you continue training the model.

Changes:

• Modified ModelTrainer._save_model to save scheduler states from active plugins
• Added scheduler_state_dict handling in Model._get_state_dict and load methods
• Updated scheduler plugins (LinearScheduler and AnnealOnPlateau) to load their states when resuming training

Closes #3444