fix: CometTakeOrderedAndProjectExec native scan node should use child operator's output

[viirya]

Which issue does this PR close?

Closes #.

Rationale for this change

What changes are included in this PR?

This bug was found while debugging CI failure in #893. In CometTakeOrderedAndProjectExec, we create internal native plan to execute native limit + sort + project. The pseudo scan node created there was incorrectly using CometTakeOrderedAndProjectExec's output attributes but it should be child node's output.

Currently it doesn't cause any error, although it will have incorrect schema in the scan node. Because sort/limit simply takes input so the incorrect schema doesn't make error. For project, we already bind attributes in Spark, as the input data is correct so it has no error too.

But in #893, we need to get the number of columns from the schema of scan node. If the schema is incorrect, the scan node will create incorrect number of array/schema structures which cause error later.

How are these changes tested?

[viirya]

Thanks @andygrove