fix(go-agent): filter SSE done sentinel from model stream (#17034)

## Summary

- Filter the `[DONE]` transport sentinel from Eino model streams
- Add regression coverage for stream messages and callbacks

## Testing

- `bash build.sh --test ./internal/entity/models`
- `bash build.sh --test ./internal/agent/component`
- `bash build.sh --test -run
'Test(AgentChatCompletions_Stream|RunAgent_Stream)' ./internal/handler`

internal/entity/models/llm.go

@@ -293,6 +293,12 @@ func (m *EinoChatModel) Stream(ctx context.Context, msgs []*schema.Message, opts
 		if content == nil && reasoning == nil {
 			return nil
 		}
+		// Provider drivers use the OpenAI-compatible [DONE] sentinel to
+		// signal the end of their transport stream. It is not assistant
+		// content and must not reach Eino's message stream or callback.
+		if content != nil && *content == "[DONE]" {
+			return nil
+		}
 		msg := &schema.Message{Role: schema.Assistant}
 		if content != nil {
 			msg.Content = *content

internal/entity/models/llm_test.go

@@ -2,11 +2,53 @@ package models
 
 import (
 	"context"
+	"errors"
+	"io"
 	"testing"
 
 	"github.com/cloudwego/eino/schema"
 )
 
+func TestEinoChatModelStreamFiltersDoneSentinel(t *testing.T) {
+	modelName := "chat"
+	driver := &streamSentinelDriver{captureToolDriver: &captureToolDriver{}}
+	var callbacks []string
+	base := NewChatModel(driver, &modelName, &APIConfig{})
+	model := NewEinoChatModel(base, &ChatConfig{
+		StreamCallback: func(content, _ string) {
+			if content != "" {
+				callbacks = append(callbacks, content)
+			}
+		},
+	})
+
+	stream, err := model.Stream(context.Background(), []*schema.Message{schema.UserMessage("hello")})
+	if err != nil {
+		t.Fatalf("Stream: %v", err)
+	}
+
+	var messages []string
+	for {
+		msg, recvErr := stream.Recv()
+		if errors.Is(recvErr, io.EOF) {
+			break
+		}
+		if recvErr != nil {
+			t.Fatalf("stream.Recv: %v", recvErr)
+		}
+		if msg != nil {
+			messages = append(messages, msg.Content)
+		}
+	}
+
+	if len(messages) != 2 || messages[0] != "answer" || messages[1] != "DONE!" {
+		t.Fatalf("stream messages = %#v, want [answer DONE!]", messages)
+	}
+	if len(callbacks) != 2 || callbacks[0] != "answer" || callbacks[1] != "DONE!" {
+		t.Fatalf("stream callbacks = %#v, want [answer DONE!]", callbacks)
+	}
+}
+
 func TestEinoChatModelGenerateSendsBoundTools(t *testing.T) {
 	apiKey := "key"
 	modelName := "chat"
@@ -148,6 +190,23 @@ type captureToolDriver struct {
 	lastConfig *ChatConfig
 }
 
+type streamSentinelDriver struct {
+	*captureToolDriver
+}
+
+func (d *streamSentinelDriver) ChatStreamlyWithSender(_ string, _ []Message, _ *APIConfig, _ *ChatConfig, sender func(*string, *string) error) error {
+	answer := "answer"
+	if err := sender(&answer, nil); err != nil {
+		return err
+	}
+	visibleDone := "DONE!"
+	if err := sender(&visibleDone, nil); err != nil {
+		return err
+	}
+	done := "[DONE]"
+	return sender(&done, nil)
+}
+
 func (d *captureToolDriver) NewInstance(baseURL map[string]string) ModelDriver { return d }
 func (d *captureToolDriver) Name() string                                      { return "capture" }
 func (d *captureToolDriver) ChatWithMessages(_ string, _ []Message, _ *APIConfig, cfg *ChatConfig) (*ChatResponse, error) {