Variable selection and causal treatment effect estimation based on interval-censored failure time data