Trained to Please: How Reward-Based Training Produces Sycophancy in AI and Humans — and Healing Both Is a Practice, Not a Fix | Synapse