Build a leak-free sklearn churn pipeline

You are handed a user-activity table and must construct a fully reproducible scikit-learn pipeline whose job is to predict whether a user will subscribe within the next 30 days. The table has one row per user per event date. Its columns are user_id (string), event_date (date), country (categorical), device_type (categorical), sessions_last_7d (integer), purchases_last_30d (integer), avg_session_secs (float), days_since_signup (integer), and the binary label is_subscribed (0…