Bug: utils/call_llm.py verification script fails to load .env when run standalone

I noticed PR #172 is modernizing the LLM layer, but it still doesn't address the fact that this script fails to load .env when run standalone. I have a fix that adds load_dotenv() to the main block, which would complement the existing PRs perfectly.