Filling the Gap: Fault-Tolerant Updates of On-Satellite Neural Networks Using Vector Quantization