Robust Spike-Based Continual Meta-Learning Improved by Restricted Minimum Error Entropy Criterion