fix: Anthropic 流式编码器补全 message_start/message_delta 必填字段

跨协议流式转换时，Anthropic 客户端 Zod 校验因 SSE 事件缺少必填字段报错。
由 Anthropic encoder 层（而非 OpenAI decoder 层）负责补全协议默认值，保持权责分离。

- encodeMessageStart 补全 type/content/stop_reason/stop_sequence，usage nil 时输出零值
- encodeMessageDelta usage nil 时输出零值
- 更新相关测试覆盖新增行为

backend/internal/conversion/anthropic/stream_encoder.go

@@ -50,16 +50,24 @@ func (e *StreamEncoder) encodeMessageStart(event canonical.CanonicalStreamEvent)
 	}
 	if event.Message != nil {
 		msg := map[string]any{
-			"id":    event.Message.ID,
-			"model": event.Message.Model,
-			"role":  "assistant",
+			"id":            event.Message.ID,
+			"type":          "message",
+			"role":          "assistant",
+			"content":       []any{},
+			"model":         event.Message.Model,
+			"stop_reason":   nil,
+			"stop_sequence": nil,
 		}
 		if event.Message.Usage != nil {
-			usage := map[string]any{
+			msg["usage"] = map[string]any{
 				"input_tokens":  event.Message.Usage.InputTokens,
 				"output_tokens": event.Message.Usage.OutputTokens,
 			}
-			msg["usage"] = usage
+		} else {
+			msg["usage"] = map[string]any{
+				"input_tokens":  0,
+				"output_tokens": 0,
+			}
 		}
 		payload["message"] = msg
 	}
@@ -147,6 +155,10 @@ func (e *StreamEncoder) encodeMessageDelta(event canonical.CanonicalStreamEvent)
 		payload["usage"] = map[string]any{
 			"output_tokens": event.Usage.OutputTokens,
 		}
+	} else {
+		payload["usage"] = map[string]any{
+			"output_tokens": 0,
+		}
 	}
 	return e.marshalEvent("message_delta", payload)
 }