Implement Model Evaluation Metrics & Explicit Train/Test Split

[BLMgithub]

Restructured the ML pipeline to separate training and holdout datasets, added model evaluation metrics, feature weights, and next-week predictions.

Implemented

• Model Evaluation: Added predictions_evaluation.sqlx and evaluation_metrics.sqlx to compute and publish forecast accuracy.
• Train/Test Dataset: Refactored ml_training_data.sqlx and ml_holdout_data.sqlx to enforce explicit temporal holdout boundaries.
• Feature Weights: Added feature_weights.sqlx to show per-feature contributions from the trained BQML model.
• Next-Week Prediction: Added predictions_nextweek.sqlx and ml_weekly_data.sqlx for next-week revenue forecasts.

Impact

• Reliability: Enforced explicit train/holdout split for reproducible out-of-sample validation.
• Observability: Added 3 views (evaluation_metrics, feature_weights, predictions_nextweek) for Dashboard.