Fix: Enable LLM-as-Judge base model evaluation integration tests, Add cleanup mechanism for MC dataset integ test

[mollyheamazon]

Issues Fixed

Integration tests for PR #5558 (LLM-as-Judge base model fix) were failing due to multiple issues:

1. Missing IAM permissions for Bedrock model invocation
2. MLflow experiment name handling causing pipeline failures
3. Display utility crashing on None evaluation scores

Changes

1. Test Infrastructure Improvements

• Enhanced error logging in integration tests to show detailed step failure information
• Removed MLflow resource ARN from test configuration (not required for base model evaluation)
• Added error handling for display utility failures to prevent test crashes

2. MLflow Experiment Name Handling

• Initially added automatic generation of mlflow_experiment_name when mlflow_resource_arn is provided

3. Display Utility Robustness

• Fixed _format_score() to handle None scores - Returns "N/A" instead of crashing with TypeError
• Prevents crashes when evaluation metrics return None (e.g., model failures, malformed outputs)

4. Dataset Cleanup

• Added cleanup mechanism for Model Customization dataset integration tests
• Simplified conftest.py fixture management

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.