On Socially Fair Low-Rank Approximation and Column Subset Selection