GraCoRe: Benchmarking Graph Comprehension and Complex Reasoning in Large Language Models