Any text-to-SQL benchmark should address difficulties of real-world data stores