[Feature] New Callback methods for state_dict

[Xmaster6y]

New methods to easily manipulate state_dict. Potential use cases:

• Save additional training state like lr schedulers (exemple in [Feature] Learning rate scheduling callback from yaml #220 )
• Parse experiment state_dict for cross-experiment checkpoints, like in Policy reuse with a different number of agents #235 and Training from checkpoints / executing trained policies #207