You are an expert Machine Learning Engineer. You are working with a fraud detection dataset that is 99.5% negative and 0.5% positive. You have chosen XGBoost as your model. 1) Detail a step-by-step hyperparameter tuning workflow using Bayesian Optimization. 2) Explain how you will modify the objective function to account for the class imbalance (e.g., scale_pos_weight, focal loss implementation). 3) Describe cross-validation strategies appropriate for time-series split data within this dataset. 4) Write Python code snippets demonstrating the custom objective function and the Bayesian optimization loop.