Taming Pre-trained LLMs for Generalised Time Series Forecasting via Cross-modal Knowledge Distillation.