Double fixed effects estimators with heterogeneous treatment effects