Explicitly map the default train split to data/train/**, use metadata.jsonl plus images/, prevent scientific validation JSON from being inferred as a split, and remove redundant embedded-image Parquet Viewer shards.
Add validated Hugging Face-native Parquet shards, point the dataset card to viewer_data/train-*.parquet, and remove the obsolete single-shard Viewer file.
