Regression in SQL support for ORDER BY and aliased expressions

[andygrove]

Describe the bug

DataFusion 10.0 supports this query:

❯ select y.column_1 z, sum(column_2) q from x y group by y.column_1 order by y.column_1;
+---+---+
| z | q |
+---+---+
| 1 | 2 |
+---+---+

DataFusion master fails:

❯ select y.column_1 z, sum(column_2) q from x y group by y.column_1 order by y.column_1;
SchemaError(FieldNotFound { qualifier: Some("y"), name: "column_1", valid_fields: Some(["z", "q"]) })

To Reproduce

echo "1,2" > data.csv
create external table x stored as csv location 'data.csv';

Expected behavior
Query should work

Additional context
None

[andygrove]

According to git bisect, this regression was introduced in #3003:

$ git bisect bad
e23925c96c97139d101719fa7a456088eeed6ae9 is the first bad commit
commit e23925c96c97139d101719fa7a456088eeed6ae9
Author: Andy Grove <[email protected]>
Date:   Mon Aug 1 19:30:26 2022 -0600

    Fix SQL planner bug when resolving columns with same name as a relation (#3003)
    
    * repro unit test
    
    * improve test
    
    * fix
    
    * make logic more robust
    
    * add a simple test case
    
    Co-authored-by: Wei-Ting Kuo <[email protected]>