Upgrade transformers at runtime for OmniVoice HiggsAudioV2TokenizerModel support

Install OmniVoice+voicetut-tts at runtime via subprocess --no-deps

Use --no-deps as global pip option to avoid conflicts with HF base image

Install OmniVoice and voicetut-tts with --no-deps to avoid pip resolution conflicts with HF base image

Pin transformers<5 and starlette<0.36 to fix HF Inference Toolkit compatibility

Add custom HF Inference Endpoint handler for VoiceTut-TTS

Duplicate from mohammedaly22/VoiceTut-TTS