auto_increment breaks binlog":
if slave's table had a higher auto_increment counter than master's (even
though all rows of the two tables were identical), then in some cases,
REPLACE and INSERT ON DUPLICATE KEY UPDATE failed to replicate
statement-based (it inserted different values on slave from on master).
write_record() contained a "thd->next_insert_id=0" to force an adjustment
of thd->next_insert_id after the update or replacement. But it is this
assigment introduced indeterminism of the statement on the slave, thus
the bug. For ON DUPLICATE, we replace that assignment by a call to
handler::adjust_next_insert_id_after_explicit_value() which is deterministic
(does not depend on slave table's autoinc counter). For REPLACE, this
assignment can simply be removed (as REPLACE can't insert a number larger
than thd->next_insert_id).
We also move a too early restore_auto_increment() down to when we really know
that we can restore the value.


mysql-test/r/rpl_insert_id.result:
  result update, without the bugfix, slave's "3 350" were "4 350".
mysql-test/t/rpl_insert_id.test:
  test for BUG#20188 "REPLACE or ON DUPLICATE KEY UPDATE in
  auto_increment breaks binlog".
  There is, in this order:
  - a test of the bug for the case of REPLACE
  - a test of basic ON DUPLICATE KEY UPDATE functionality which was not
  tested before
  - a test of the bug for the case of ON DUPLICATE KEY UPDATE
sql/handler.cc:
  the adjustment of next_insert_id if inserting a big explicit value, is
  moved to a separate method to be used elsewhere.
sql/handler.h:
  see handler.cc
sql/sql_insert.cc:
  restore_auto_increment() means "I know I won't use this autogenerated
  autoincrement value, you are free to reuse it for next row". But we were
  calling restore_auto_increment() in the case of REPLACE: if write_row() fails
  inserting the row, we don't know that we won't use the value, as we are going to
  try again by doing internally an UPDATE of the existing row, or a DELETE
  of the existing row and then an INSERT. So I move restore_auto_increment()
  further down, when we know for sure we failed all possibilities for the row.
  Additionally, in case of REPLACE, we don't need to reset THD::next_insert_id:
  the value of thd->next_insert_id will be suitable for the next row.
  In case of ON DUPLICATE KEY UPDATE, resetting thd->next_insert_id is also
  wrong (breaks statement-based binlog), but cannot simply be removed, as
  thd->next_insert_id must be adjusted if the explicit value exceeds it.
  We now do the adjustment by calling
  handler::adjust_next_insert_id_after_explicit_value() (which, contrary to
  thd->next_insert_id=0, does not depend on the slave table's autoinc counter,
  and so is deterministic).