Investigating causality in associations between education and smoking: A two-sample Mendelian randomization study