KRISTEVA: Close Reading as a Novel Task for Benchmarking Interpretive Reasoning