On the optimum sequential test of two hypotheses for statistically dependent observations