Add vi back via fairseq model_name + TTS_HOME mirror approach

Remove vi from Space (fairseq config incompatible with coqui-tts 0.27.x)

Fix nested model_args.speakers_file path in config patch

Overwrite config blob in-place to fix absolute speaker path

Fix config.json absolute path rewriting for HF Space

Initial Space: FastAPI + Coqui TTS for en/es/vi