`SubprocessCluster` and `SSHCluster` hang indefinitely when using the new-style logging configuration

When using the new-style logging configuration, both the `SSHCluster` and the `SubprocessCluster` hang indefinitely because they rely on a stream handler registered to `distributed` on `INFO` level to write the scheduler address to `stderr`.

This can be reproduced by using new-style logging configuration in your Dask configuration file, e.g. your config may look like
```yaml
distributed:
version: 2
logging:
version: 1
```
and starting a cluster.

With the old-style logging configuration this works as expected due to #8391.