How would you structure a data lake to support both batch processing and ad-hoc SQL queries at Robinhood?