The rapid proliferation of large language models (LLMs) is creating new opportunities for automated code generation in integrated development environments (IDEs). However, the effectiveness of these systems depends not only on the underlying LLMs but also significantly on how contextual information is represented, selected, and integrated within real-world software development workflows. In this paper, we conduct a systematic review of context representation techniques for LLM-based code generation, with a focus on IDE integration scenarios, particularly the contextual signals available in Visual Studio environments. The study follows the PRISMA methodology and analyzes 51 selected papers from reputable academic sources. Based on the synthesized findings, we propose a multidimensional taxonomy for context-aware code generation and identify key challenges related to context extraction, output reliability, human-machine interaction, and system evaluation. Finally, we outline promising research directions for developing context-aware code generation systems that are both efficient and practical for real-world IDE deployment.