Implicit to Explicit Entropy Regularization: Benchmarking ViT Fine-tuning under Noisy Labels