It looks like samll revolution in SELECT_LEX tree, but it was only natural way to solve problem with name resolution of external fields inside subselect which belongs to global order of union
also it have following advantages:
 - removed mess with current_select type conversion
   - type checking/converting
   - a lot of virtual methods
 - fake select for union execution allocated only once (it was allocated for every subselect with union executing)

changes:
fixed bug with outer fields name resolution of subqueries which belong to global ORDER BY clause
remuved select_lex() function, now thd->lex.current_select always have type SELECT_LEX
new SELECT_LEX (fake_select_lex) will be allocated in case of UNION for using in UNION processing
fake_select_lex allocated for union hold global ORDER BY & LIMIT clauses and have linkage equal to GLOBAL_OPTIONS_TYPE 
new description of SELECT_LEX tree

(SCRUM)