gpt-5.4 returns `arguments` as a JSON object for some tools (e.g.
apply_patch, tool_search), but ResponsesOutput.Arguments was typed as
string, causing json.Unmarshal to fail on every such SSE event.

The failure silently dropped all output_item events and the
response.completed event from the forwarded stream, so:
- the client never received tool call content
- usage/token counts could not be extracted (billed as 0)
- client reported "stream closed before response.completed"

Fix: change Arguments to json.RawMessage and add GetArguments() which
normalises both forms to a plain string (unescapes JSON strings,
returns raw bytes for objects).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

This reverts commit ff14caed7c.

- ClaudeRequest 新增 CacheControl 字段
- RequestOpenAI2ClaudeMessage 自动注入顶层 cache_control: {"type":"ephemeral"}
- System 复合内容和普通消息 content block 的 CacheControl 透传
- ParseContent 新增 []MediaContent 类型断言，修复透传路径
- 新增 8 个单元测试覆盖自动注入、透传、序列化等场景

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- codex adaptor 支持 API key 和 OAuth 两种认证模式
- 提取 setupOAuthHeader 和 shouldUseChatCompletionsViaResponses
- 修复编辑页 codex_credential_mode 切换/回显不同步问题
- ResponsesStreamResponse 添加 Error 字段支持独立错误事件

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

从 QuantumNous/new-api 上游移植以下修复：

- fix(ssrf): 加固视频代理和 MJ 图片代理的 SSRF 防护，FetchModels 提权至 RootAuth
- fix(claude): 流式断流时不再整份覆盖 usage，保留 cache 计费字段
- fix(pricing): 定价接口增加用户可用分组过滤
- fix(ratio): compact 模型优先匹配精确定价再回退通配符
- fix(ali): 接受 task polling 返回的字符串类型 usage 值
- fix(gemini): 从 URL 路径 :streamGenerateContent 检测流式请求
- fix(zhipu_4v): coding plan 图片生成使用正确的 OpenAI endpoint
- fix(openai): MessageImageUrl.Detail 字段添加 omitempty
- fix(channel-affinity): skip_retry_on_failure 配置项为 false 时不再丢失
- fix(document-renderer): 使用 useMemo 替代 useState 处理 HTML payload
- fix(fetch-setting): ApplyIPFilterForDomain 默认值改为 true

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>