KVM: X86: implement the logic for spinlock optimization

get_cpl requires vcpu_load, so we must cache the result (whether the
vcpu was preempted when its cpl=0) in kvm_vcpu_arch.

Signed-off-by: Longpeng(Mike) <longpeng2@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>