A Hardware-Aware Performance Analysis of Machine Learning Models for Sensor-Aided Millimeter-Wave Beam Prediction