Currently the schema of the Object table in the output parquet files depend on what filters are included in the input dataset. For example the Object table in a parquet file from the ci_hsc dataset has fewer columns than one from the HSC-RC2 dataset, because the former contains 2 filters and the latter contains 5. It is not ideal that the schema is not fixed. We would like the output schema to be deterministic and independent of the input dataset for the same camera (or obs package). Perhaps a config parameter in the obs package would be suitable.
It's okay to only consider obs_subaru for now.