On Post-selection Inference in A/B Testing