Add end-to-end tutorial with real-world dataset

[gauthierpiarrette]

The quickstart uses synthetic data. A tutorial using a real (public) dataset would help users understand how Timefence fits into an actual ML workflow.

Candidates:

• IEEE-CIS Fraud Detection (Kaggle) - transactions with timestamps
• Instacart Market Basket - orders with temporal ordering
• Any open feature-store-style dataset with event timestamps

The tutorial should cover: loading data, defining sources/features, building, auditing, and interpreting the report.