With the switch to tracing we did not set the `max_log_level` anymore.
This resulted in a performance degradation as logging did not early
exited and thus, `trace` logs were at least resolved every time.
This pr fixes it by ensuring that we set the correct max log level.