fix: openenv.yaml tasks graders + strict score bounds

fix: add safe_reward() clamp at every reward return point

Minimal patch: standalone proxy ping + reward clamped to (0,1)

Add normalize_reward(); enforce strict (0,1) on all paths

Update grading logic to strict (0,1) ratio fallback

fix: use API_BASE_URL/API_KEY for LiteLLM proxy — always make API call (Phase 2)

fix: make inference.py crash-proof when OPENAI_API_KEY is missing (Phase 2)

fix: OpenEnv multi-mode compliance — add main() entrypoint and uv.lock

Add pyproject.toml for OpenEnv validation compliance

Production-ready: add server/app.py with fallback-safe /reset, fix Dockerfile, add HF metadata, add task JSON files

Initial commit