Align inference with hackathon validator

Normalize exported score/reward fields to strict (0,1)

Ensure fallback inference output always includes 3 valid tasks

Handle inference exceptions and avoid strict-mode crashes

Use validator proxy API_BASE_URL+API_KEY in strict inference mode

fix: rename 'final_score' to 'score' in inference output payload

fix(critical): rewrite inference.py to use in-process environment

fix: clamp evidence_score and all boundary score values to strict (0,1) range

fix(eval): strictly bound scores within 0.001 and 0.999 to resolve hackathon validation out-of-range limits

fix(inference): Provide valid Pydantic model for exception fallback and ensure structural logs are printed

fix(inference): handle Unhandled Exceptions and connect to standard environment API

Hackathon submission: Guardian OpenEnv