fix: clamp reward scores to strict (0, 1) range

fix: implement abstract methods in SQLAnalyticsEnv and use sync() wrapper

fix: rewrite inference.py to match official sample key=value log format

fix: pass api_key to OpenAI client from OPENAI_API_KEY env var

fix: add uv.lock and server entry point in project.scripts

fix: add HF_TOKEN and LOCAL_IMAGE_NAME env vars to inference.py per checklist

fix: add HuggingFace Space metadata with app_port 8000

feat: SQL Analytics OpenEnv environment — 10 tasks, dense rewards, RL-ready