perf: implement benchmark-driven optimizations

fix: suppress Whisper duplicate logits-processor warning; add gTTS 429 retry backoff

Full audit fixes: src/__init__.py, tmp TTS, soundfile+accelerate, remove HF_TOKEN warning, clean imports

Clean rewrite: whisper-small local, clean RAG+LLM with Hindi prompts

Upgrade: Use openai/whisper-large-v3 via Inference API for CPU compatibility

Clean commit without venv