Gravar-mail: Tradeoff between moving targets, gradient magnitude and performance in quantum variational Q-Learning