Bug #25093 slave replay of transaction with CREATE TMP table fails

Description:
Given the transaction executing by slave
  begin;
       ...
       CREATE temporary table;
       update that deadlocks or triggers timout;
  commit;
which gets deadlocked or timed-out and replayed
`CREATE temp' executes for the second time.
This fails because of temp was created at a previous round.

How to repeat:
connection master;
create table t1 (a int) engine=innodb;
insert into t1 values (0);
create table t2 (a int) engine=innodb;
insert into t2 values (0);

slave_sync_with_master;
connection slave;
begin;
      insert into t2 values (1),(1),(1);
      update t2 set a=2;

connection master;
begin;
       update t1 set a=1;
       CREATE temporary table tt (a int);
       update t2 set a=1;
commit;

connection slave;
      update t1 set a=2 /* deadlock, replicated ta victim */;

061215 14:47:16 [ERROR] Slave: Error 'Table 'tt' already exists' on query. Default database: 'test'. Query: 'CREATE temporary table tt (a int)', Error_code: 1050
061215 14:47:16 [Warning] Slave: Table 'tt' already exists Error_code: 1050

Suggested fix:
tolate the error of `table alread exists' on slave.
It's safe:
 slave knows it replays, and that's a temporary table which exists.

Andrei,
*if* the replication team implements Mats's suggestion of making a temporary table created inside a transaction to be dropped at rollback, the present bug would disappear (because, when the slave thread gets a timeout it does a rollback of the entire transaction, and that would drop the temporary table)