DINGO: Distributed Newton-Type Method for Gradient-Norm Optimization