In redis#9408, we added some SENTINEL DEBUG to reduce default
timeouts and allow tests to execute faster. The change
in 05-manual.tcl may cause a race that SENTINEL FAILOVER
response with a NOGOODSLAVE:
```
Manual failover works: FAILED: Expected NOGOODSLAVE No suitable replica to promote eq "OK" (context: type eval line 6 cmd {assert {$reply eq "OK"}} proc ::test)
(Jumping to next unit after error)
FAILED: caught an error in the test
assertion:Expected NOGOODSLAVE No suitable replica to promote eq "OK" (context: type eval line 6 cmd {assert {$reply eq "OK"}} proc ::test)
```

The reason is that the info-period value was reduced in redis#9408
(the default value is 10000), and then manual failover was
performed immediately, but the INFO may not exchanged between
the sentinel and replicas, causing the sentinel to skip all
the replicas in sentinelSelectSlave (Because replica's info_refresh
is not updated, see the code snippet below), then return a NOGOODSLAVE,
break the test.

Code snippet from sentinelSelectSlave:
```
while((de = dictNext(di)) != NULL) {
    sentinelRedisInstance *slave = dictGetVal(de);
    mstime_t info_validity_time;
    if (master->flags & SRI_S_DOWN)
        info_validity_time = sentinel_ping_period*5;
    else
        info_validity_time = sentinel_info_period*3;
    if (mstime() - slave->info_refresh > info_validity_time) continue;
}
```

By adding a wait_for_condition, we have the opportunity to
let sentinel update the info_period of the replicas.