Resplit dataset with 6:4 train/test ratio while maintaining subject_code distribution

Add Multilingual Poetry Subject Clustering Benchmark - 题材 Clustering Benchmark

initial commit