docs(comments): 注释中文化 · internal/runtime/multimodal（21 文件，纯替换）

- 只翻注释/文档字符串，代码零变化：go/scanner 词法流比对 21 文件 token+字面量全等，增删行数 1:1（467/467）
- 刻意保留非散文英文：JSON Schema description、工具 Description() 返回值、adapter prompt、API 路径与命令示例、环境变量名
- 逐行确认代码前缀未变，未触碰其他模块

internal/runtime/multimodal/vision/vision.go

@@ -1,15 +1,15 @@
-// Package vision implements the layered image routing described in
-// DevDocs/design/06-多模态.md §0.
+// Package vision 实现 DevDocs/design/06-多模态.md §0 所述的
+// 分层图像路由。
 //
-// A request that carries image content is first checked by DetectMultimodal.
-// The VisionRouter then picks one of the layered routes: the main model
-// directly (when it accepts images), a tool-shaped vision/OCR adapter (when the
-// main model self-assesses poorly or cannot see), or the OCR tool (when the
-// image is mainly text). Vision/OCR adapters are registered as sdk.Tool
-// capabilities through the capability registry, and their output is a tool
-// result, so the image itself never occupies main-model context.
+// 携带图像内容的请求先由 DetectMultimodal 检查。
+// 随后 VisionRouter 选出其中一条分层路由：主模型
+// 直连（当它接受图像时）、工具形态的 vision/OCR 适配器（当主模型
+// 自评较差或看不见图像时），或 OCR 工具（当
+// 图像以文字为主时）。vision/OCR 适配器经能力注册中心
+// 注册为 sdk.Tool 能力，其输出即工具
+// 结果，因此图像本身从不占用主模型上下文。
 //
-// Images may come from a local path, an http(s) URL, or a base64/data: URI.
+// 图像可来自本地路径、http(s) URL 或 base64/data: URI。
 package vision
 
 import (
@@ -18,11 +18,11 @@ import (
 	"git.yeij.top/AskaEth/Cyrene_v2/internal/runtime/model"
 )
 
-// DetectMultimodal reports whether msgs contains any non-text content part,
-// i.e. whether multimodal handling is required. Today model only distinguishes
-// text from image content, so this is true exactly when a message carries an
-// image part; the check is written against "not text" so future content types
-// (audio) extend it without rework.
+// DetectMultimodal 报告 msgs 是否含任何非文本内容部件，
+// 即是否需要多模态处理。当前 model 只区分
+// 文本与图像内容，因此这等价于某条消息携带
+// 图像部件；判断写成「非文本」是为了让未来的内容类型
+// （音频）无需返工即可扩展。
 func DetectMultimodal(msgs []model.Message) bool {
 	for _, m := range msgs {
 		for _, part := range m.Content {
@@ -34,30 +34,30 @@ func DetectMultimodal(msgs []model.Message) bool {
 	return false
 }
 
-// ImageSource classifies where an image's bytes come from.
+// ImageSource 归类一副图像的字节来自何处。
 type ImageSource string
 
 const (
-	// SourcePath is a local filesystem path.
+	// SourcePath 是本地文件系统路径。
 	SourcePath ImageSource = "path"
-	// SourceURL is an http(s):// URL.
+	// SourceURL 是 http(s):// URL。
 	SourceURL ImageSource = "url"
-	// SourceBase64 is a base64 payload, optionally wrapped in a data: URI.
+	// SourceBase64 是 base64 载荷，可选地包在 data: URI 中。
 	SourceBase64 ImageSource = "base64"
 )
 
-// ImageRef is a single image reference plus its resolved source kind.
+// ImageRef 是单个图像引用及其已归类的来源种类。
 type ImageRef struct {
-	// Source is the classified source kind.
+	// Source 是归类后的来源种类。
 	Source ImageSource
-	// Data is the local path, URL, or base64/data: payload.
+	// Data 是本地路径、URL 或 base64/data: 载荷。
 	Data string
 }
 
-// ParseImageRef classifies a raw image source string into an ImageRef. Empty
-// input is rejected. Local paths that exist on disk are classified as paths;
-// http(s) URLs as URLs; every other value, including a data: URI, is treated as
-// a base64 payload. This keeps routing deterministic without guessing MIME.
+// ParseImageRef 把原始图像来源字符串归类为 ImageRef。空
+// 输入被拒绝。磁盘上存在的本地路径归类为路径；
+// http(s) URL 归类为 URL；其余所有值（含 data: URI）都视为
+// base64 载荷。这样无需猜测 MIME 即可保持路由确定性。
 func ParseImageRef(src string) (ImageRef, error) {
 	if src == "" {
 		return ImageRef{}, ErrNoImage
@@ -68,7 +68,7 @@ func ParseImageRef(src string) (ImageRef, error) {
 	if strings.HasPrefix(src, "file://") {
 		return ImageRef{Source: SourcePath, Data: strings.TrimPrefix(src, "file://")}, nil
 	}
-	// A data: URI is base64 already; treat as base64, not path.
+	// data: URI 本身就是 base64；按 base64 处理，而非路径。
 	if strings.HasPrefix(src, "data:") {
 		return ImageRef{Source: SourceBase64, Data: src}, nil
 	}