Introduce Visit::visit_select

[xitep]

• SetExpr is a directly recursive data-structure through SetExpr::SetOperation
• Its visit impl dispatches directly to Select (without necessary going through Query first)
• Right now a Visitor impl has no chance to observe when a SELECT starts and ends during the traversal

The motivation is that SELECTs often introduce aliases which are valid only for that particular SELECT.

select * 
  from (select 1 val from dual) x
 where x.val = 1

 union all

select *
  from (select 2 val from dual) y
 where y.val = 2

-- referring to `x` at this point is an error (at least on oracle)
-- referring to `x` or `y` as part of an `order by` is an error (at least on oracle)

This change allows effectively to determine the scope of a (nested) SELECT using Visitor/Mut.

Alternatively we could drop the SetExpr::Select variant and replace it entirely with the existing SetExpr::Query; after all, a Select is a Query's body. (My personal preference, however, would be to keep the current structure as it makes the AST easier to follow / navigate / more expressive.)

• related issue: Question: why is the Visitor trait limited to statements, relations & expressions? #934

[iffyio]

LGTM! Thanks @xitep!