Causal Inference on Missing Exposure via Triple Robust Estimator