refactor: adjust score clamping range to 0.001-0.999 and improve action normalization robustness

Fix inference.py: sync heuristic fallback, clamp all scores--2

Fix inference.py: sync heuristic fallback, clamp all scores

Fix #19: clamp overall score in inference.py and update baseline_results.json

Checklist Phase II --- part 3

Checklist Phase II --- part 2

Fix Phase 2: clamp all grader scores to (0.0001, 0.9999) exclusive

Checklist Phase II

Fix Phase 2 validator log formats

Update variables and logs for final checklist submission

refactor: update baseline results and inference logic for improved model performance metrics

Rename baseline_inference.py to inference.py for OpenEnv checks