With the development of increasingly interconnected cyber-physical systems (CPSs), the Internet of Things (IoT) paradigm must be expanded further to account for the collection, transmission, and processing of unprecedented amounts of data in uncertain and changing environments. Cognitive Internet of Things (CIoT) introduces a paradigm shift in IoT systems by integrating the engineering perspective of cognition, as formulated in cognitive dynamic systems (CDS), into traditional IoT frameworks. This survey systematically examines how CIoT leverages the five pillars of cognition: perception, attention, memory, language, and intelligence, to enable context-aware, autonomous, and adaptive functionality. We trace the evolution from standard IoT architectures to this cognitively enriched model, detailing how data acquisition and storage, combined with enabling technologies such as data fusion, reinforcement learning, cognitive communications (via cognitive radios), and the integration of foundation models and large language models (LLMs), facilitate advanced data analytics and introduce a new intelligent layer for deeper contextual understanding and adaptation. By emphasizing the synergy between CDS principles and emerging technologies, the paper demonstrates how CIoT can address longstanding challenges in scalability, interoperability, and resource management. Through a critical evaluation of current limitations and lessons learned, we offer a forward-looking perspective on how these cognitively inspired frameworks can further enhance intelligent IoT ecosystems. Ultimately, this work serves as a foundational resource for aligning IoT systems with the engineering-driven notion of cognition, guiding future research and innovation in autonomous, scalable IoT environments.