Too Brittle to Touch: Comparing the Stability of Quantization and Distillation towards Developing Low-Resource MT Models