fix: ensure task graders are discoverable for phase 2 checks

update task difficulty calibration and validation coverage

fix: clamp rewards to 0.01-0.99 and add Grader classes

Improve grading: add keyword aliases for flexible description matching

Improve grading matching and inference with Qwen 32B

Quality snippets: 27 real-world cases (Django, Flask, React, Express, TypeScript, security bugs)

Add 38 more snippets: 12 Python style, 12 bug hunt (JS/Py/TS), 22 security (Py/JS/TS). Total: 62 snippets

Merge: Use ported code review environment

Initial commit: Code Review Environment for Scaler Meta PyTorch Hackathon

Finalize OpenEnv hackathon submission