Deep neural networks algorithms for stochastic control problems on finite horizon: Convergence analysis