Enhancing Low-Resource NMT with a Multilingual Encoder and Knowledge Distillation: A Case Study