In #175 / #138 we improved support for GeoParquet writing; however, we can't add the bbox to the key/value metadata without copying a huge chunk of the Parquet writer into SedonaDB and so I didn't implement that as a first step (the workaround is to write GeoParquet 1.1 which allows readers to better prune row groups anyway).
Also related is #133 (support for Parquet GEOMETRY/GEOGRAPHY) which is orthogonal to this particular issue but we should tackle first because it affects the code we need to copy. We can tackle that issue when we upgrade to DataFusion 51.