Can Input Attributions Interpret the Inductive Reasoning Process Elicited in In-Context Learning?