Spin off from SOLR-6261.
This kind of already happens because the only session watcher in ConnectionManager does it's processing async (changed in SOLR-5615), but this is more consistent and avoids the possibility that a second session watcher or a change to that code re-surfaces the issue again.