Can we convert dictionary-encoded parquet columns to Categorical Dtype?

[delucchi-cmu]

For something like band, we know that there are only a few possible values, and parquet knows this too and does a dictionary-style encoding on those columns. Can we consider using this hint to convert those columns to Categorical Dtype and save space in-memory?

Is there an in-memory savings to be gained?

[hombit]

It may be related to pyarrow's dictionary type:
https://arrow.apache.org/docs/python/generated/pyarrow.dictionary.html
I think it could be already supported by us if pandas supports it?