Fixed a number of memory leaks discovered by valgrind.

dbug/dbug.c:
  This is actually an addendum to the fix for bug #52629:
  
  - there is no point in limiting the fix to just global
  variables, session ones are also affected.
  - zero all fields when allocating a new 'state' structure so
  that FreeState() does not deal with unitialized data later.
  - add a check for a NULL pointer in DBUGCloseFile()
mysql-test/r/partition_error.result:
  Added a test case for bug #56709.
mysql-test/r/variables_debug.result:
  Added a test case for bug #56709.
mysql-test/t/partition_error.test:
  Added a test case for bug #56709.
mysql-test/t/variables_debug.test:
  Added a test case for bug #56709.
sql/item_timefunc.cc:
  There is no point in declaring 'value' as a member of
  Item_extract and dynamically allocating memory for it in
  Item_extract::fix_length_and_dec(), since this string is only
  used as a temporary storage in Item_extract::val_int().
sql/item_timefunc.h:
  Removed 'value' from the Item_extract class definition.
sql/sql_load.cc:
  - we may need to deallocate 'buffer' even when 'error' is
    non-zero in some cases, since 'error' is public, and there is
    external code modifying it.
  - assign NULL to buffer when deallocating it so that we don't
    do it twice in the destructor
  - there is no point in changing 'error' in the destructor.