Machine learning based prediction of optical properties in two-dimensional Mo-W-S-Se-Te transition-metal dichalcogenide alloys through physics-informed sampling