sched_ext: Remove unnecessary cpu_relax()

As described in commit b07996c7 ("sched_ext: Don't hold
scx_tasks_lock for too long"), we're doing a cond_resched() every 32
calls to scx_task_iter_next() to avoid RCU and other stalls. That commit
also added a cpu_relax() to the codepath where we drop and reacquire the
lock, but as Waiman described in [0], cpu_relax() should only be
necessary in busy loops to avoid pounding on a cacheline (or to allow a
hypertwin to more fully utilize a core).

Let's remove the unnecessary cpu_relax().

[0]: https://lore.kernel.org/all/35b3889b-904a-4d26-981f-c8aa1557a7c7@redhat.com/



Cc: Waiman Long <llong@redhat.com>
Signed-off-by: David Vernet <void@manifault.com>
Signed-off-by: Tejun Heo <tj@kernel.org>