Better error reporting in mysql-test-run (don't hide the important message

with 'head').
Test fix (this sync_with_master was nonsense).
Now all tests pass in 4.0 in autobuild.


mysql-test/mysql-test-run.sh:
  Error reporting (how mysql-test-run filters the messages which mysqltest prints
  on stderr) is not reliable. It does a 'head the first line only' which in
  my case caused me to lose some time, because in fact the worrying message was
  on the second line (logically, the one which caused the die() is at the
  end of the $TIMEFILE file, not at the beginning).
  Doing a 'tail' is better but not perfect (depending on the build, one may get
  "Maximum memory usage" as the last line); it's always good to have all messages
  and spot the important one oneself. So we print all $TIMEFILE (that's a few lines
  at most).
mysql-test/t/rpl_rotate_logs.test:
  Dont sync_with_master as the SQL slave thread may be stopped at this moment;
  just wait_for_slave_to_stop. Doing sync_with_master caused a non-deterministic
  behaviour, because (note that save_master_pos was called to early) sometimes
  MASTER_POS_WAIT() started before the SQL thread died (because of unique
  violation) so returned 0, other times it started after the SQL thread died
  so returned NULL (which is reported by mysqltest as an error, on stderr).

mysql-test/mysql-test-run.sh

@@ -569,7 +569,9 @@ error () {
 }
 
 error_is () {
-    $CAT < $TIMEFILE | $SED -e 's/.* At line \(.*\)\: \(.*\)/   \>\> Error at line \1: \2<\</' | $HEAD -1
+    $ECHO "Errors are (from $TIMEFILE) :"
+    $CAT < $TIMEFILE
+    $ECHO "(the last line(s) may be the ones that caused the die() in mysqltest)"
 }
 
 prefix_to_8() {

mysql-test/t/rpl_rotate_logs.test

@@ -73,12 +73,9 @@ insert into t2 values(1234);
 
 #same value on the master
 connection master;
-save_master_pos;
 set insert_id=1234;
 insert into t2 values(NULL);
 connection slave;
-sync_with_master;
-
 wait_for_slave_to_stop;
 
 #restart slave skipping one event