[FEA] Update cudf.pandas Parquet reading according to cudf-polars results

Based on our findings in https://github.com/rapidsai/cudf/pull/16944 we may want to update cudf.pandas to parametrize the chunked parquet reading accordingly, in particular by setting a new pass read limit. The exact value may need to be different based on 1) cudf.pandas generally targeting smaller GPUs and 2) cudf.pandas using UVM, so we may have to do some extra research and testing there.