[Feature]Support structured reward outputs and grouped reward aggregation for multi-reward RL workflows