Serverless-safe defaults: fallback from LLaMA/TinyLlama to DialoGPT; 404 auto-fallback to distilgpt2; robust prompt/cleanup for both families

Fix: switch default model from gpt2 to Llama-3.2-1B; add Llama prompt/cleanup; update models list; add test scripts

Remove HTML interface, keep API only

Replace app.py with pure API version - no HTML interfaces