prompt_text parameter in ONNX mode

[GergelyZsolt]

It would be very useful if the prompt_text parameter could be specified in onnx mode as well, as it would improve the quality of the model. If possible, it would also be useful if it could be specified in voice_cloning mode as well.

[alpacaking]

For the current released MOSS-TTS-Nano model, the supported inference workflow is voice_clone inference. The public web demo and the ONNX demo are both designed around this voice-clone workflow, and the reference audio transcript / prompt_text is not used as an input in this mode.

The continuation mode in the PyTorch inference code is mainly provided for models that have been SFT-trained with continuation-mode data, so that such fine-tuned models can be used for inference. It is not the official/default inference mode for the currently released model.