fix(record): persist L1 extraction failure diagnostics

[RerankerGuo]

Description | 描述

Persist L1 extraction failure diagnostics when the extraction model returns an empty or unparsable response.

• Writes .metadata/l1-extraction-failures.jsonl with failure reason, parse message, session identifiers, source message ids, model, and a capped raw LLM response.
• Keeps the existing extraction semantics unchanged: failed parses still produce zero extracted memories, but operators can now inspect what the model returned after content filtering or malformed JSON.
• Adds a focused regression test for malformed extraction output.

Related Issue | 关联 Issue

Fix #104

Change Type | 修改类型

• Bug fix | Bug 修复
• New feature | 新功能
• Documentation update | 文档更新
• Code optimization | 代码优化

Self-test Checklist | 自测清单

• Verified locally | 本地验证通过
• No existing features affected | 无影响现有功能

Additional Notes | 其他说明

Verified with npx vitest run src/core/record/l1-extractor.test.ts, npm test, and npm run build using Node v24.15.0.