Effective Off-Policy Evaluation and Learning in Contextual Combinatorial Bandits