[Bug]: Qwen3-TTS, when set load_format:dummy in qwen3-tts.yaml, ValueError: This checkpoint does not provide `speaker_encoder` weights; cannot compute ref_spk_embedding from ref_audio.