Improve handling of hybrid v1/v2 cgroups

Description

related:

• https://github.com/microsoft/WSL/issues/12711
• https://github.com/microsoft/WSL/issues/13006
• https://github.com/docker/for-win/issues/14801
• https://github.com/docker/for-win/issues/14704

The daemon currently accounts for the host to be either cgroup v1 or v2, but doesn't properly handle situations where the host is running hybrid v1/v2.
https://github.com/moby/moby/blob/829b6953756f1ad21769f318ec9717ed854936b3/pkg/sysinfo/sysinfo_linux.go#L88-L95

On hosts where (some) ome v1 controllers are in use (hierarchy != 0), resulting in the daemon failing to start;

failed to start daemon: Devices cgroup isn't mounted

On such hosts;

 cat bad/proc-cgroups.utf8
#subsys_name    hierarchy       num_cgroups     enabled
cpuset  0       57      1
cpu     2       4       1   <- this
cpuacct 0       57      1
blkio   0       57      1
memory  1       4       1   <- this
devices 0       57      1
freezer 0       57      1
net_cls 0       57      1
perf_event      0       57      1
net_prio        0       57      1
hugetlb 0       57      1
pids    0       57      1
rdma    0       57      1

Start with pkg/sysinfo/sysinfo_linux.go around lines 88-95 and compare its cgroup detection with the provided hybrid /proc/cgroups output. Reproduce the daemon startup failure on a host using v1 and v2 controllers, then verify that hybrid cgroup hosts no longer fail with “Devices cgroup isn't mounted”.