Remove the redundant copy of join key column

[gengliqi]

Enhancement

tiflash/dbms/src/Flash/Coprocessor/DAGExpressionAnalyzer.cpp

Lines 1246 to 1258 in 961f9dd

	if (!has_actions && (!left || is_right_out_join))
	{
	/// if the join key is a columnRef, then add a new column as the join key if needed.
	/// In ClickHouse, the columns returned by join are: join_keys, left_columns and right_columns
	/// where left_columns and right_columns don't include the join keys if they are ColumnRef
	/// In TiDB, the columns returned by join are left_columns, right_columns, if the join keys
	/// are ColumnRef, they will be included in both left_columns and right_columns
	/// E.g, for table t1(id, value), t2(id, value) and query select * from t1 join t2 on t1.id = t2.id
	/// In ClickHouse, it returns id,t1_value,t2_value
	/// In TiDB, it returns t1_id,t1_value,t2_id,t2_value
	/// So in order to make the join compatible with TiDB, if the join key is a columnRef, for inner/left
	/// join, add a new key for right join key, for right join, add new key for both left and right join key
	String updated_key_name = unique_name_generator.toUniqueName((left ? "_l_k_" : "_r_k_") + key_name);

As the comments above said, the schema returned by join in ClickHouse is different from TiDB. To make them consistent, TiFlash copied the join key column for the left side(if it's right outer join) and right side.
Although this copy is a shadow copy, it introduced unnecessary logic and made the code complex.
We can change the schema of join and remove the redundant copy of join key column because it's not difficult.