Default log handler doesn't work when a global hazelcast_client instance is used [API-2364]

The default log handler uses a [static local mutex ](https://github.com/hazelcast/hazelcast-cpp-client/blob/12aa0dc6f9a6ef0b376053ad54171d2eaaa6ce95/hazelcast/src/hazelcast/logger.cpp#L99)for synchronizing access to `std::cout`. If a global `hazelcast_client` instance is used in user code and the shutdown() method isn't called before `main()` ends, the dtor of hazelcast_client will implicitly call shutdown and try to print logs during the process. As the order in which the mutex and hazelcast_client be destroyed is not guaranteed, the client may end up accessing an already destroyed mutex.

To reproduce:
```c++
std::vector clis;

int main() {
clis.push_back( hazelcast::new_client().get() );
}
```

The exception thrown from std::mutex::lock will cause the program to abort during termination on macOS.