Trainability Issues in Quantum Policy Gradients with Softmax Activations Artigo de Conferência