Explain the trade-offs between quantization, pruning, and knowledge distillation for edge inference in Syntiant-style on-device models.
