[BUG] GpuSequence blows up when nulls exist in any of the inputs (start, stop, step).

[firestarman]

Describe the bug
GpuSequence blows up when nulls exist in any of the inputs (start, stop, step),

Caused by: ai.rapids.cudf.CudfException: cuDF failure at: ../src/lists/sequences.cu:148: steps input column must not have nulls.
  at ai.rapids.cudf.ColumnVector.sequences(Native Method)
  at ai.rapids.cudf.ColumnVector.sequence(ColumnVector.java:533)

but Spark just returns nulls.

+---+---+----------------------+
| _1| _2|sequence(_1, _2, NULL)|
+---+---+----------------------+
|  1|  2|                  null|
|  3|  4|                  null|
+---+---+----------------------+

Steps/Code to reproduce bug
Launch a Spark shell with GPU support, and run
Seq((1, 2), (3, 4)).toDF.repartition(1).selectExpr("*", "sequence(_1, _2, null)").show

Expected behavior
We should keep align with what Spark does for null cases.

[firestarman]

Filed the FEA issue #10012 to cudf for this

[firestarman]

Close due to wrong project. Here is the right one NVIDIA/spark-rapids#4499