Estimating long-term causal effects from short-term experiments and long-term observational data with unobserved confounding