This is not an unrecoverable situation.  Users of kvm_read_guest_offset_cached
and kvm_write_guest_offset_cached must expect the read/write to fail, and
therefore it is possible to just return early with an error value.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>