CodeBrain: Towards Decoupled Interpretability and Multi-Scale Architecture for EEG Foundation Model