Extending Graph Condensation to Multi-Label Datasets: A Benchmark Study