cpuset: let hotplug propagation work wait for task attaching

Instead of triggering propagation work in cpuset_attach(), we make
hotplug propagation work wait until there's no task attaching in
progress.

IMO this is more robust. We won't see empty masks in cpuset_attach().

Also it's a preparation for removing propagation work. Without asynchronous
propagation we can't call move_tasks_in_empty_cpuset() in cpuset_attach(),
because otherwise we'll deadlock on cgroup_mutex.

tj: typo fixes.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>