fix(record): harden L1 parsing and embedding timeout

[RerankerGuo]

Description | 描述

Harden the L1 extraction path against non-strict LLM JSON output and ensure capture-path embedding timeout configuration reaches vector dual-write calls.

• Parse the first valid scene payload instead of greedily matching from the first [ to the last ], so explanatory text like [姓名]）在 [时间] no longer poisons extraction.
• Accept object-wrapped scene arrays such as { "scenes": [...] } while still ignoring unrelated JSON before the actual scene payload.
• Forward embedding.captureTimeoutMs / capture embedding timeout through writeMemory() so slow CPU embedding providers are not limited by the default embedding timeout during L1 vector writes.

Related Issue | 关联 Issue

Fix #110

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 npm test and npm run build using Node v24.15.0.