Learning with Dual-level Noisy Correspondence for Multi-modal Entity Alignment