Noise-Robust Target-Speaker Voice Activity Detection Through Self-Supervised Pretraining