Pre-wire instruct support in generate_voice_clone for upcoming 25Hz model

[ezzaldeeen]

Summary

• Added optional instruct parameter to generate_voice_clone, allowing users to pass natural-language style instructions alongside voice cloning.
• Follows the same instruct_ids pattern already used in generate_voice_design and generate_custom_voice.
• Fully backwards compatible, no changes required for existing callers.

Note

⚠️ This feature is experimental and may produce unstable results.
The 12Hz base model was not trained for instruct-guided voice cloning. As noted in the README, instruction control in the base model is currently unstable. Full support for combined cloning + instruction is planned for the upcoming 25Hz voice editing model.
Use at your own discretion and treat outputs as experimental.

It's a very important feature for us. Thanks @wangxiongts and your team for your great work. Feel free to drop any feedback/comments if needed.