[FEA] Make `max_file_samples` and `max_rg_samples` configurable in cudf-polars

[rjzamora]

#19276 introduced functionality for sampling statistics from both Parquet metadata and row-group data. For now, he maximum number of files and row-groups used to sample statistics are called max_file_samples and max_rg_samples, respectively.

These parameters should be user configurable. Some users may want more-accurate statistics, while others may want to avoid gathering any statistics at all.

Note: We probably want to make this change after #19369 gets in.