fix: Optimized and support for inference context return

[chenyme]

Summary

「回传推理上下文」自 5781834b 起完全失效：reasoning_content 正常入库，但从未被读出来过。

根因是 ListMessageAncestors / ListMessageAncestorsUntil 两处 WITH RECURSIVE CTE 手写了列清单并漏掉 reasoning_content。该清单原本是为规避「GORM Scan 遇到未知字段会出错」而写的——这个前提经实测是错的，GORM 按列名映射并静默忽略未匹配列（gorm/scan.go:276-281，与驱动无关）。所以本 PR 不只补列，而是把两处外层投影改为 SELECT *，从根上消除「手写列清单随新增列漂移」这一 bug class。

排查中发现同一 bug class 的另外两处，一并修复：cloneSharedMessage 手写字段清单同样漏掉 ReasoningContent（且复制了 ReasoningTokens，造成行内自相矛盾）；中断生成的落库更新未带该字段，而 UpdateAssistantMessageCompletion 是无条件覆盖，会把已产出的推理抹成空——interrupted 状态的 assistant 消息仍会进入后续轮次上下文。

另外扩展了 Chat Completions 的回传厂商白名单，加入 moonshot / zhipu（issue 中提到的 K3 需求）。两家均要求历史 assistant 消息原样携带 reasoning_content：Moonshot 缺失时服务端直接返回 400，智谱默认 clear_thinking=false 即保留思考。字段名与放置位置和现有序列化一致，未改动协议层。

Fixes #529

Change type

• Bug fix
• Feature
• Documentation
• Refactor
• Configuration / deployment
• Security hardening
• Other

Affected areas

• Frontend / UI
• Backend / API
• Authentication / authorization
• Conversations / streaming
• Files / RAG / extraction
• Model routing / providers
• MCP / tools
• Billing / payments
• Admin console
• Deployment / Docker / configuration
• Documentation

Verification

• pnpm check — 通过（api-contract 契约一致、biome lint 544 文件无问题、前后端 typecheck）
• pnpm test — 通过
• go build ./... / go vet ./... / gofmt -l internal/ — 全部干净
• go test ./... — 55 包全绿，0 失败

补充验证：

• 变异测试：每处修复都撤销后确认对应测试变红，再恢复，确保守卫不是「写完就绿」
• 真实 Postgres 18：CTE 改动在临时容器中跑通真实 repo 方法（此前仅覆盖 SQLite in-memory），确认 SELECT * 的语法与 _depth 辅助列扫描行为一致

未验证：没有挂真实 DeepSeek / Kimi / GLM 渠道端到端抓包确认出站请求携带 reasoning_content。建议合并后在同一对话连发 3 轮实测确认。

Screenshots, API examples, or logs

变异测试输出（撤销修复时）：

ListMessageAncestors dropped columns
ListMessageAncestors: ancestor 1 reasoning content = "", want "reasoning 2"
cloned reasoning content = "", want preserved
expected trimmed reasoning to be retained, got "stale"
ancestor walked into conversation 1: ... Content:"FOREIGN_SECRET" ...

真实 Postgres 18 验证：

postgres OK: reasoning="PG_HISTORICAL_REASONING" pricing="{"in":1}" reasoningTokens=125

Configuration, migration, and compatibility notes

• 无数据库迁移：reasoning_content 列早已由迁移添加（postgres.go:397），本 PR 仅修正读取
• 无配置变更，无环境变量、Docker、CORS、公开 URL 变动
• API 契约未变（pnpm check 的 swagger/TS 契约校验通过），无生成物改动
• 前端零改动：回传开关是全局布尔（chat.reasoning_content_passback），vendor 判定全在后端；开关文案本就是 vendor 无关的「支持的 Chat Completions 推理模型」
• 向后兼容：SELECT * 对 DB 多出的列忽略、模型多出的字段取零值，行为不弱于原显式清单
• PostgreSQL 与 SQLite 均受影响且均已修复——两者共用 internal/infra/persistence/postgres 仓储实现（persistence/sqlite/ 只含连接配置）

Documentation

• Documentation is not needed for this change.
• Documentation was updated.
• Documentation still needs to be updated.

Security and privacy

• No secrets, tokens, credentials, local config, or personal data are included.
• User data access remains scoped by authenticated user context unless an admin-only path explicitly requires broader access.
• Security-sensitive behavior was reviewed, including authentication, authorization, provider routing, file processing, billing, and admin APIs where relevant.

安全加固说明：ListMessageAncestors 的递归项此前未约束 m.conversation_id（兄弟方法 ListMessageAncestorsUntil 有）。经核查不是当前可利用漏洞——三条写入路径均强制同会话父指针，HTTP 层虽接受调用方传 parentMessagePublicID，但经会话+用户双重限定的 GetMessageByPublicID 校验后才落库。但 parent_message_id 上没有外键，不变量纯靠应用层保证；且本 PR 的 SELECT * 会扩大一旦不变量被破坏时的泄漏面（原本连推理内容都读不出来）。已补上约束并加回归测试，测试绕过应用层直接植入跨会话父指针，验证遍历会停在会话边界。

Checklist

• I searched existing issues and pull requests.
• Changes are focused and do not include unrelated refactors.
• Tests or static verification were run where practical.
• User-facing behavior, deployment steps, API contracts, or configuration changes are documented.
• Generated artifacts are included only when this project explicitly requires them.
• Caches, build output, .pyc files, .env files, and local storage data are not committed.