The case for an iceberg-native database