Description of problem:
A machine with an invalid CPU set - out of the range of available host cpus - starts successfully but is expected to fail. This seemingly happens inconsistently on a specific host with a specific number of CPUs and doesn't reproduce on other archs or s390x hosts where two different error messages are displayed and the machines wouldn't start (as expected).

Version-Release number of selected component (if applicable):
libvirt-8.9.0-2.el9.s390x
kernel-5.14.0-197.el9.s390x

How reproducible:
100% on specific s390x host

Steps to Reproduce:
1. On a host with 12 CPUs (value "CPU(s)" from lscpu|grep CPU), define a domain with cpu set out of range, e.g.
<vcpu cpuset="0-12,^2" current="3" placement="static">8</vcpu>
or equivalently
<vcpu cpuset="0-1,3-12" current="3" placement="static">8</vcpu>

2. Try to start the vm

Actual results:
The VM starts successfully

Expected results:
The VM doesn't start, instead there's an error message from cgroups "Numerical result out of range"

Additional info:
A) Yingshun confirmed the expected result on x86_64.

B) In the reported case we see cgroup values

1. cat /sys/fs/cgroup/machine.slice/machine-qemu\\x2d1\\x2davocado\\x2dvt
   x2dvm1.scope/libvirt/emulator/cpuset.cpus
   0-1,3-12
2. cat /sys/fs/cgroup/machine.slice/machine-qemu\\x2d1\\x2davocado\\x2dvt
   x2dvm1.scope/libvirt/emulator/cpuset.cpus.effective
   0-1,3-11

C) On another s390x host with 4 CPUs and the same component versions, this can't be reproduced, instead another error message is showing. With
<vcpu placement='static' cpuset='0-4,^2' current='3'>4</vcpu>
we get an error message
error: Invalid value '0-1,3-4' for 'cpuset.cpus': Invalid argument

Auto case is vcpu_affinity.negative_test.vcpu.outrange_cpuset