Add requirements section and pandas dtype_backend example

Polars example: aggregate remotely, collect small results

Lighter DuckDB example

Fix usage examples: polars hf bug workaround, robust pyarrow glob

Move manifest and checksums to repo root so data/ holds only Parquet

Add Zenodo DOI to card

Add 35 annual Parquet files (67,257,982 records, Jul 1983 - Jun 2017)

Add metadata specification

Add data description

Add dataset card

initial commit