Set temperature to 0.0 for LLM-as-judge in evaluate_risk.py and evaluate_summaries.py

[harpomaxx]

Problem

Both evaluate_risk.py and evaluate_summaries.py use temperature=0.3 when calling the judge LLM. This introduces randomness into the evaluation, making results non-deterministic and harder to reproduce.

Fix

Set temperature=0.0 in call_judge_llm() in both scripts to ensure deterministic, reproducible evaluations.