Full-fabric VHDL LLM inference engine. Runs Qwen3.5-class transformer inference