Functional Benchmarks for Robust Evaluation of Reasoning Performance, and the Reasoning Gap