Update read_parquet `dtype_backend` description to be in line with the new pandas description

[dougbrn]

https://github.com/lincc-frameworks/nested-pandas/blob/main/src/nested_pandas/nestedframe/io.py#L53

https://pandas.pydata.org/docs/dev/reference/api/pandas.read_parquet.html#pandas.read_parquet

[hombit]

I'd disallow dtype_backend other than pyarrow (with NestedDtype applied automatically) in our version of the function. Especially with our implementation of the nested serialization proposal