Wrong type conversion for columns of DaskDataFrame in Xenium

[LucaMarconato]

Working on a xenium notebook I noticed this:

xe_rep1_sdata.points['transcripts']['feature_name'].cat.as_known().compute()

gives

0                              b'RUNX1'
1                              b'ITGAX'
2                               b'TCIM'
3                               b'TCIM'
4                                b'LUM'
                        ...            
126857385                      b'HMGA1'
126857386                      b'CLIC6'
126857387                       b'SQLE'
126857388                        b'DST'
126857389    b'NegControlCodeword_0529'
Name: feature_name, Length: 126857390, dtype: category
Categories (541, object): ['b'ABCC11'', 'b'ACTA2'', 'b'ACTG2'', 'b'ADAM9'', ..., 'b'WARS'', 'b'ZEB1'', 'b'ZEB2'', 'b'ZNF562'']

I haven't checked, but I suspect that the xenium() function in this repo converts the categories from bytes object to strings. We should call .decode('utf-8') or something similar instead.

The bug could also be in the io when writing/reading to/from parquet. We need to check.

[LucaMarconato]

The problem is in this conversion to str of a columns containing bytes representation (from the points parser):

if feature_key is not None:
    feature_categ = dd.from_pandas(
        data[feature_key].astype(str).astype("category"),
        **kwargs,
    )  # type: ignore[attr-defined]
    table[feature_key] = feature_categ

I will fix this by converting decoding the bytes in spatialdata-io (in _get_points() from xenium.py), so that the parser receives strings already.