Datasets and descriptors

xtalforge_analysis::Dataset is a JSON Lines container: DatasetRecord { id, structure, properties }, stable DATASET_VERSION, and to_json_lines/from_json_lines plus save/load. column and property_names expose tabular properties; iteration preserves file order.

deduplicate(dataset, &DedupOptions) uses the structure matcher with a composition/supercell prefilter and returns a greedy DedupReport with the kept and removed groups. The report's apply returns a duplicate-free dataset.

composition_features, rdf_descriptor, coulomb_matrix and soap_like_descriptor return numeric descriptors for ML. The SOAP-like descriptor is a Gaussian radial basis times Legendre angular moments and is documented as an approximation, not exact SOAP.

Python exposes the same container as xtalforge.Dataset with JSONL I/O and optional pandas/polars views.

Parity keys: dataset, dataset_dedup, ml_descriptors.