hack: fall back when cgroup remount setup is unavailable

[crazy-max]

relates to:

    #1 [internal] booting buildkit
    #1 pulling image moby/buildkit:buildx-stable-1
    #1 pulling image moby/buildkit:buildx-stable-1 8.5s done
    #1 creating container buildx_buildkit_fervent_perlman0
    #1 29.74 creating cgroup namespace
    #1 creating container buildx_buildkit_fervent_perlman0 21.2s done
    #1 29.74 creating cgroup namespace
    #1 29.74 �������>umount: can't unmount /sys/fcreating cgroup namespace
    #1 29.74 umount: can't unmount /sys/fs/cgroup: Operation not permitted
    #1 29.74 umount: can't unmount /sysumount: can't unmount /sys/fs/cgroup: Operation not permitted
    #1 29.74 creating cgroup namespace
    #1 29.74 creating cgroup namespace
    #1 29.74 �������>umount: can't unmount /sys/fcreating cgroup namespace
    #1 29.74 umount: can't unmount /sys/fs/cgroup: Operation not permitted
    #1 29.74 umount: can't unmount /sysumount: can't unmount /sys/fs/cgroup: Operation not permitted
    #1 29.74

#6368 made the entrypoint automatically attempt the cgroup namespace remount workaround whenever the current cgroup path is not /. That works for the Kubernetes case, but it might break in other cases where unshare succeeds but the helper then fails to umount /sys/fs/cgroup with "Operation not permitted", which leaves the BuildKit container restarting during bootstrap in Buildx.

cc @marxarelli

[ctalledo]

LGTM ...