Traning-free Uncertainty-guided Logit Adjustment for Few-shot Class-Incremental Learning