A No-Defense Defense Against Gradient-Based Adversarial Attacks on ML-NIDS: Is Less More?