Attempting to write pl.Categorical to Parquet

cat.py

from string import ascii_uppercase

import numpy as np
import polars as pl

N = 5  # N > 4 to break to_parquet/read_parquet


# for smaller N, categories need to be multiple characters
cats = [a + b for a in ascii_uppercase for b in ascii_uppercase][:N]

df = pl.DataFrame(
    {
        "cats": pl.Series(cats).cast(pl.datatypes.Categorical),
        "numbers": np.arange(N),
    }
)

df.to_parquet("df.parquet", use_pyarrow=True)  # this works

_ = pl.read_parquet("df.parquet")

df.drop("cats").to_parquet("df.parquet") # this does too

_ = pl.read_parquet("df.parquet")

df.with_column(df.cats.cast(pl.datatypes.Utf8))\
    .to_parquet("df.parquet") # so does this

_ = pl.read_parquet("df.parquet")

df.to_parquet("df.parquet", use_pyarrow=False)  # this doesn't

_ = pl.read_parquet("df.parquet")