[FEA] Support mapInArrow introduced by pyspark 3.3.0+

[wbo4958]

Spark 3.3.0 has introduced a new API mapInArrow in PySpark DataFrame, see SPARK-37228 and PR apache/spark#34505. mapInArrow is quite similar with mapInPandas, the only difference is the input is Iterable[pa.RecordBatch] for mapInArrow, while it is Iterator[pd.DataFrame] for mapInPandas.

PyArrow has already supported CUDA Integration, see https://arrow.apache.org/docs/python/integration/cuda.html and potential CUDA IPC, which means, there is a chance that Rapids Accelerator has the opportunity to support ZERO-COPY between JVM process an python process and improve the performance.

I hope it can be supported in Spark-Rapids in 22.12 release.