Add ROC-AUC score to compute metrics function in ISSR_AI4MH_Yixing_Fan/pipeline/test_model.py

[agentksimha]

Added roc_auc_score to compute_metrics using softmax probabilities (class 1) rather than hard predictions, which is required for a valid AUC calculation.

Justification:

• if Dataset is likely class-imbalanced; AUC is more reliable than accuracy alone
• Captures model performance across all decision thresholds, useful for tuning the classification cutoff in a safety-critical context
• Aligns with standard evaluation practice for binary NLP classifiers