fix: 会话摘要包含用户 prompt + 更新 CLAUDE.md + 升级 SDK

[lishuceo]

Summary

• 会话摘要从只保存 output 末尾改为 用户: <prompt前200字> | 回复: <output末300字>，修复 agent 跨 thread 时不知道用户之前问了什么的问题
• CLAUDE.md 更新：补充 vitest、新增模块描述、Agent SDK Gotchas section
• SDK 升级 0.2.42 → 0.2.45

Test plan

• Build 通过
• 飞书实测新摘要格式

🤖 Generated with Claude Code

[claude]

Claude finished @lishuceo's task in 2m 49s —— View job

---

Code Review for PR #25

• Check for previous review comments — none found
• Gather context and read changed files
• Review code changes
• Post review summary

---

✅ Approved — Clean, focused PR with no issues found.

Changes reviewed:

• src/feishu/event-handler.ts — Summary format changed from output-only (last 500 chars) to 用户: <prompt≤200> | 回复: <output last 300>. The prompt parameter is typed string, always defined at call site. Logic is correct.
• src/pipeline/runner.ts — Same summary format change applied consistently. prompt is destructured from the pipeline store record and is always a non-null string.
• CLAUDE.md — Documentation updates: vitest command, new module descriptions, Agent SDK gotchas section, workspace env vars. All accurate and well-organized.
• package.json / package-lock.json — SDK bump 0.2.42 → 0.2.45, clean lockfile update.

Assessment: The summary format change is a good improvement — including the user prompt gives the agent cross-thread context it previously lacked. The total summary length is bounded (date prefix + dir + 200 char prompt + 300 char tail ≈ ~600 chars max), which is reasonable. Both call sites apply the identical logic, keeping behavior consistent. No bugs, security issues, or architectural concerns.