Patch 1 - fix crash (part 2 of problem)

This problem appears to have two parts:
 1. When the stored procedure/prepared statement is executed two times,
    it is optimized differently on the second execution, and
 2. When the first execution uses semi-join materialization, and the
    second does not, a crash during optimization may occur.

This patch handles the crash that occurs during second execution, a
later patch will handle the different optimization.

The root cause of the crash is that a dangling pointer is left after
first execution: The sj_mat_exec field of TABLE_LIST for a semi-join nest.
The solution is a slight refactoring, so that the TABLE_LIST no longer
refers Semijoin_mat_exec, instead the Semijoin_mat_exec contains a
pointer to the TABLE_LIST object. At the same time the definition of
Semijoin_mat_exec is moved from sql_class.h into sql_executor.h
(it is an execution struct).