Currently the GeoParquet reader ignores the "crs" key of a column on read:
SedonaDB does some minimal parsing of the PROJJSON to extract this information, which is usually embedded in the JSON:
https://github.com/apache/sedona-db/blob/e6cd9c391b2ed6a50464abb17a972f8d4ce8ed9d/rust/sedona-schema/src/crs.rs#L274-L287
Of course, if the CRS is omitted, the SRID can be hard-coded to 4326 because that is the CRS of a GeoParquet file.
This SRID is propagated to GeoPandas via dataframe_to_arrow() if included, which would make the GeoParquet reader in Sedona more useful in an arbitrary context.