Composable Coresets for Determinant Maximization: Greedy is Almost Optimal