Addressing '行号幻觉' (line number hallucination) in LLM-based code review.

This issue exposes a critical technical limitation in LLM-driven code review: '行号幻觉' (line number hallucination). The user, also developing an LLM CodeReview tool, identifies the core problem as the LLM's '空行理解不准确' (inaccurate understanding of empty lines) and proposes a solution involving 'Diff文件补上行号标识' (adding line number identifiers to Diff files). This directly challenges the 'precise line-level comments' claim in the repo context. The current '复杂' (complex) and '稳定性不够' (insufficient stability) solution for line number positioning indicates a significant engineering hurdle. Resolving this hallucination is paramount for maintaining developer trust and the tool's credibility, especially in 'battle-tested at Alibaba's scale' environments where accuracy is non-negotiable. This highlights a fundamental accuracy and reliability pain point in current LLM-based code analysis.