fix: handle empty choices from LLM response (list index out of range)

fix: switch to Qwen2.5-72B-Instruct (available on HF free tier)

fix: use same-origin API calls in production + HEAD method support for Next.js prefetch

fix: langchain_text_splitters import for LangChain v0.3+

feat: HF Spaces Docker deployment — non-root user, model pre-download, keep-alive timeout

Changes Pro max