What is the problem the feature request solves?

Just tried comet locally with spark-shell, it turns out that comet requires that all the input shall be columnar based. So, it's not possible to test comet with the following code:

spark.range(10, 1000, 10).selectExpr("id", "id + 1 as val").repartition(10, col("id"))

Adding the CometRowToColumnar operator on top of the leaf node(RangeExec) will make it easy to test comet out.

In the long term, I think the CometRowToColumnar could be used to wrap around Spark's row-base source exec node and makes all the following operators columnar-based and leverages comet's columnar execution.

Describe the potential solution

Add CometRowToColumnar operator

Additional context

No response