TuringQ: Benchmarking AI Comprehension in Theory of Computation