Fix dataset metadata for load_dataset

Update dataset loading examples and single-file split paths

Consolidate each dataset split into one Parquet file

Publish FactGuard-Bench with train, validation, and test splits

initial commit