Tools to benchmark expression performance

### Description

As gluten has more and more native expressions support, it's better to have a separate tool to stress/test expressions only.
The general idea is:
- using pyarrow to generate some mock data, with respect to the required data type. e.g., valida json, DATE format string
- using pyspark to call the target spark functions
- collect the output, write into parquet/noop format, so spark can iterate the whole records

-yuan

The issue names no files or tests. Start by mapping the proposed pyarrow mock-data generation, PySpark function invocation, and Parquet/no-op output flow; done would be a separate tool that stress-tests expressions across the required data types.

Written by the indexing model from the issue text.