Sparse Orthogonal Parameters Tuning for Continual Learning