An Embarrassingly Simple Baseline for Imbalanced Semi-Supervised Learning