[Bug] Anthropic API Error: Reasoning Extraction False Positives in Extended Sessions

**Bug Description**
"관측용 기술조사 세션에서 [reasoning_extraction] 연쇄 오탐" Request ID: req_011CeDop1tgtMDVEnoJytNt5, req_011CeDomHsnkYmAgmdum19Lx, req_011CeDogDxFdWeYuxwek6x2r, req_011CeDoeYvU1o8Lw8QJcKxNF, req_011CeDocLDRuGqM5LDuJmgri, req_011CeDoaKQWNBrTvSixLf8R3, req_011CeDoUDMD6kkvNrYjKSMBa, req_011CeDoRAyzPpAinkLcTq2rW, req_011CeDoJTrJfNX4sYyC3grqR, req_011CeDnYeptF65cK2iqLGRHR, req_011CeDnRuRmwt513hHxSVZG7, req_011CeDnNxbSMiZeRwVbUMhrH, req_011CeDkzXVbtzArWRKMrVtXm, req_011CeDkjpBGxcPVjN7wsj8tC, req_011CeDfYAdg7Ki6yfGfDpMHh, req_011CeDfnST4vAUKY9FG2ZBCc,

**Environment Info**
- Platform: win32
- Terminal: null
- Version: 2.1.237
- Feedback ID: b8c6a7f2-15ce-4913-aa89-bacb194c3727

**Errors**
```json
[]
```

Start by reproducing the reported reasoning_extraction false positives on win32 with version 2.1.237, using the listed request IDs and feedback ID as investigation references. The report names no files, tests, or entry points; completion would require identifying why extended sessions trigger false positives and verifying that the behavior no longer occurs.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.