Towards Sequence-to-Sequence Reinforcement Learning for Constraint Solving with Constraint-Based Local Search