[CALCITE-6691] QUALIFY on project references wrong columns

The bug was that we were translating the QUALIFY expression after
the SELECT clause, and the mapping was wrong if columns had moved
around of it the expressions were non-trivial. The fix is to convert
the QUALIFY expression at the same time as the HAVING expression.

Also add support for a QUALIFY clause in a query with a GROUP BY. In
this case, the QUALIFY expression may reference aggregate functions.
Previously such queries would give a ClassCastException (trying to
convert a LogicalAggregate to a LogicalProject).

If a QUALIFY expression references or duplicates an expression in the
SELECT clause, we no longer detect and deduplicate that. This has
made one or two plans more verbose. Potentially we would add back
deduplication. Or just let Calc do common subexpression elimination
later in the planning process, as it always has done.

Also refactor SqlToRelConverter. Previously we were creating some
`SqlNodeList` wrappers only to use their `accept(SqlVisitor)` method;
now we leave them as `List<SqlNode>` and call new method
`SqlVisitor.visitAll(List<SqlNode>)`.

Close #4061

core/src/main/java/org/apache/calcite/sql/util/SqlVisitor.java

@@ -26,6 +26,8 @@
 import org.apache.calcite.sql.SqlNodeList;
 import org.apache.calcite.sql.SqlOperator;
 
+import java.util.List;
+
 /**
  * Visitor class, follows the
  * {@link org.apache.calcite.util.Glossary#VISITOR_PATTERN visitor pattern}.
@@ -103,4 +105,9 @@ public interface SqlVisitor<R> {
   default R visitNode(SqlNode n) {
     return n.accept(this);
   }
+
+  /** Visits all nodes in a list. */
+  default void visitAll(List<SqlNode> selectList) {
+    selectList.forEach(e -> e.accept(this));
+  }
 }