Steerable Multi-Objective Group-Relative Preference Optimization with Jointly Trained Low-Rank Adapters