Silently truncation occurs in analysis phase of ingestion

Summary

The analysis phase of ingestion has a fixed cap of 4096 token. It turns out that it may be insufficient for ingesting long and complicated raw documents. However, the app now does not have any logging to remind user of truncation. It simply carries on ingestion with truncated analysis from LLM.

Environment

App Version: 0.6.11 OS: Windows 11 model: deepseek-v4-flash

Reproduce

Just need to ingests file in attachment. The significant lack of pages generated from ingestion implies the problem.

Fix

I add a warning to remind user. To really fix this issue, it may need a dynamic elevation of 4096 cap.

孔乙己.txt