Neural-network-based finite-time H∞ control for extended Markov jump nonlinear systems