EXHIB: A Benchmark for Realistic and Diverse Evaluation of Function Similarity in the Wild