This patch is for 5.7.

Issue 1
=======

A quoted identifier name consisting of a single asterisk character
could recognized as an unquoted asterisk, i.e.

  SELECT `*` ...

was parsed as

  SELECT * ...

Issue 2
=======

A qualified quoted identifier with a name starting with the asterisk character
could recognized as as unquoted asterisk if the same select list contained an
unquoted asterisk symbol, i.e.

  SELECT a.`*`, ... b.* ...

was parsed as

  SELECT a.*, ... b.`*` ...

Analysis for issue 1
====================

After the WL7200, a special processing of qualified [<schema>].table.*
and unqualified * in a select list has diverged: the unqualified one has
went to Item_field::itemize(), but the qualified has become PTI_table_wild.

As well as it's impossible to mistake quoted and unquoted asterisk inside
PTI_table_wild, it is also impossible (too late) to distinguish quoted
and unquoted names inside Item_field::itemize().
Thus, the code for the unqualified asterisk in Item_field::itemize() has
failed to recognize identifiers like a back-quoted `*` as identifiers.

Analysis for issue 2
====================

From the beginning, the resolver (see setup_wild()) has been designed to
recognize asterisks in a select list with a help of:

  1. an Item_field placeholder objects with "*" in place of their column names,

  2. a counter of asterisk select list items in SELECT_LIST::with_wild local
     to each query block,

Thus, a loop in setup_wild() has searched over all select list items
checking all Item_field objects (including either placeholders for * or
regular column reference items) for the 1st character of Item_field::field_name
until it reached SELECT_LIST::with_wild of such objects.

So, if a select list contained unquoted asterisk items before column names
containing "*" at their start, everything worked fine.
However, if there was a column name starting with "*" before an unquoted
asterisk in a select list, the former recognized as an unquoted asterisk,
while the latest recognized as a quoted column name.

Solution
========

1. The code branch in Item_field::itemize() that processed asterisks has
   been removed.

2. The Item_asterisk class has been introduced to replace PTI_table_wild.
   Reasoning:

     * unlike PTI_table_wild, that disappears after the
       PTI_table_wild::itemize() call leaving Item_field in
       its place, the new Item_asterisk class can survive till
       the setup_wild() call to be distinguished there from
       other Item_field objects those represent regular
       column references.

     * minus one MEM_ROOT allocation

     * the new name Item_asterisk is closer to the standard

3. All direct allocations of Item_field for the unquoted asterisk thing
   have been replaced with allocations of Item_asterisk.

4. setup_wild() has been updated to recognize asterisk items using
   Item_field::is_asterisk() instead of looking into Item_field::field_name.

Change-Id: Id0e5eea03608ed6140901e32d840dc3190a67bee