Testing the Limits of SMILES-based De Novo Molecular Generation with Curriculum and Deep Reinforcement Learning