CLEME2.0: Towards More Interpretable Evaluation by Disentangling Edits for Grammatical Error Correction