Learning-Enhanced Safeguard Control for High-Relative-Degree Systems: Robust Optimization under Disturbances and Faults