cgroup, sched: implement PRIO_RGRP for {set|get}priority()

One of the missing features in cgroup v2 is the ability to control cpu
cycle distribution hierarchically among threads of a process.  With
rgroup infrastructure in place, this can be implemented as a natural
extension of setpriority().

This patch introduces a new @which selector PRIO_RGRP for
{set|get}priority() which can be used only when the calling thread is
in a rgroup and respectively sets and gets the nice priority of the
rgroup that the calling thread belongs to.  The nice values have
exactly the same meaning as for a single task and top-level rgroups
compete with peer tasks as if the entire subtree is a single task with
the specified nice value.

setpriority(PRIO_RGRP, nice) automatically enables cpu controller upto
the rgroup of the thread.  The cpu controller is available iff it's
mounted on the default hierarchy and available on the nearest sgroup
(ie. the parent of the nearest sgroup should have it enabled in its
subtree_control).  If the controller isn't available, setpriority()
fails with -ENODEV.

If the cpu controller is made unavailable either through clearing of
subtree_control or migration to a cgroup which doesn't have it
available, cpu controller is disabled for the affected rgroup
subtrees.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paul Turner <pjt@google.com>