Document the `__index_level_0__` parquet trick

[fmigneault]

If a Parquet file is encoded its index happening to be an int64 representation of the DGGS "index", it is possible to configure as such to save space and let pydggsapi do the inline conversion.

    "h3-collection":
      classname: "parquet_collection_provider.ParquetCollectionProvider"
      datasources:
        "h3-collection":
          filepath: "path/to/h3_int64.parquet"
          id_col: __index_level_0__  # "h3"
          data_cols: [...]

By "using the int64 index", I mean as in:

hex_population_canada_df['h3'] = hex_population_canada_df['h3'].apply(lambda zone_id: int(zone_id, 16))

as shown in https://github.com/crim-ca/ogc-dggs/blob/main/canada-population/data_preparation.ipynb