Qwen3-TTS Silently Dropped My Instruction, and My A/B Never Noticed

In qwen_tts 0.1.1, generate_custom_voice sets the instruction to None on the 0.6B model without any warning, so half of my speech-rate A/B compared runs that never received it. Before comparing settings of a control knob, show that it moves a measurable output such as words per minute or F0.