Prediction of Reynolds stresses in high-Mach-number turbulent boundary layers using physics-informed machine learning