Currently the calcite projection also flattens the output row in PlannerImpl.
This causes failure when inserting into a table with a ROW field.
E.g. INSERT INTO table (f_row_destination) SELECT f_row_source will not work, as f_row_source will be flattened into separate fields and won't match the destination schema
Imported from Jira BEAM-4514. Original Jira may contain additional context.
Reported by: kedin.