1. Merge inference_webui and inference_gui so that the GUI no longer needs to reload models for each inference, while also allowing the use of either WebUI or GUI to be determined by the 'USE_WEBUI' environment variable.
2. Simplify GUI's code and address various inefficiencies, including: enabling direct input of ref text and target text (akin to the WebUI), facilitating file selection for ref audio uploads, adding language options for CH-EN/JA-EN/Multi (with Multi as the default), standardizing variable name to enhance readability.
