Tuning for Trustworthiness -- Balancing Performance and Explanation Consistency in Neural Network Optimization