Error occurred when transforming from row to a new row without setCoder

The output row from transform function uses the input row schema to verify which causes error. Ex:

```

.apply(MapElements.via(
new SimpleFunction() \{

@Override
public Row apply(Row line) {
return Row.withSchema(newSchema).addValues("a",
1, "b").build();
}
}));

```

Got error when the output row schema is not the same as the input row.

Need to add `.setCoder(RowCoder.of(newSchema))` after the transform function to make it work.

Related link: [https://stackoverflow.com/questions/61236972/beam-sql-udf-to-split-one-column-into-multiple-columns](https://stackoverflow.com/questions/61236972/beam-sql-udf-to-split-one-column-into-multiple-columns)

Imported from Jira [BEAM-9814](https://issues.apache.org/jira/browse/BEAM-9814). Original Jira may contain additional context.
Reported by: ruixuel.