Empty point extraction job results should not be downloaded

[kvantricht]

We have quite a few empty parquet files resulting from point extraction jobs. The reason for this is that currently we have to buffer point-based extractions because occasional single-point jobs are not yet supported in openEO (Open-EO/openeo-geopyspark-driver#996)

Very small buffers sometimes do not contain any pixel center of a collection and the resulting extraction is nodata for all observations. These get dropped, finally resulting in an empty dataframe.

Things to do:

• Change post-job action to detect empty dataframes and don't download these (they give downstream issues)
• Make sure current point-extractions have square 5m buffer for the actual extraction, avoiding this getting dropped
• waiting for final fix in openEO, after which none of the above should be essential anymore (load_stac: support Point features Open-EO/openeo-geopyspark-driver#996)

[kvantricht]

Should be addressed by #266