fix: update grader scores to fall strictly within (0, 1)

Fix KeyError: MODEL_NAME and add global error handling

Enforce strict usage of LLM proxy environment variables

Standardize API environment variables, update port to 7860, and bump version to 0.2.0

Prepare and optimize for Hugging Face Spaces deployment

feat: add submission-compliant inference.py with OpenAI client