GTPO: Stabilizing Group Relative Policy Optimization via Gradient and Entropy Control