Add maskable GraphPPO based on sb3_contrib.MaskablePPO + GNN for domains with graph observations

[nhuet]

• Derive MaskableGraphPPO from MaskablePPO (and also GraphOnPolicyAlgorithm to reuse work done for GraphPPO)
• Derive MaskableGNNActorCriticPolicy from MaskableActorCriticPolicy (and similarly from _BaseGNNActorCriticPolicy to share code with GNNActorCriticPolicy), idem for MaskableMultiInputGNNActorCriticPolicy.
• Update stable_baselines3 scikit-decide wrapper with a new argument use_action_masking in order to wrap the domain in an environment exposing action_masks method (redirecting towards domain.get_action_mask())
• We add examples for GraphMaze and GraphJspDomain so that the solver propose only applicable actions