Enhancement: Support categorical read/write with TabularDatasets

Native pandas preserves categorical data types when writing and reading to Parquet format. Unfortunately, Azure's TabularDataset implementation does not preserve this nice behavior.

This is more of a nice to have than a requirement, but I could imagine cases with (say) GBM models which accept categoricals not being able to leverage TabularDatasets due to this behavior.

There is a sample notebook to reproduce the issue here: https://github.com/vla6/Azure_notes/blob/main/tabulardataset_parquet_categoricals_issue_di.ipynb