Fixed three bugs that made Qwen3.5-122B a daily driver on Mac Studio

A developer switched from DS4 Flash to Qwen 3.5-122B on their M3 Mac Studio after discovering that long-context prefill latency (3-5 minutes for follow-up questions beyond 50k tokens) made pair programming impractical. They subsequently fixed three bugs in their serving stack unrelated to the model choice that were necessary to make the new model usable for their long-context agentic coding workflow.