Fix the chained UDF with Row type

[elvaliuliuliu]

This is a follow PR of #376 which will fix the following chained UDF issue with Row type (#407):

    # Chained UDFs with GenericRow type.
    {
	var schema = new StructType(new[]
	{
	    new StructField("col1", new IntegerType()),
	    new StructField("col2", new StringType())
	});
	Func<Column, Column> udf1 = Udf<string>(
	    str => new GenericRow(new object[] { 1, "abc" }), schema);

	Func<Column, Column> udf2 = Udf<Row, string>(
	    row => row.GetAs<string>(1));

	Row[] rows = _df.Select(udf2(udf1(_df["name"]))).Collect().ToArray();
	Assert.Equal(3, rows.Length);

	var expected = new[] { "abc", "abc", "abc" };
	string[] actual = rows.Select(x => x[0].ToString()).ToArray();
	Assert.Equal(expected, actual);
    }

We compared with PySpark UDFs and found that Schema is not passed in chained UDF cases. So schema-less Row is created to handle chained UDF.

[imback82]

there is no conversion happening here.

[elvaliuliuliu]

This constructor is for converting GenericRow to schema-less Row, right?

[imback82]

well, it's just setting the given value to its member (not "convert"ing it)

[elvaliuliuliu]

Got it, make sense.

[imback82]

LGTM, thanks!