Architectural Backdoors in Deep Learning: A Survey of Vulnerabilities, Detection