feat(generation): migrate article generation and template-assist to RabbitMQ queues

Replace in-process generation with queue-based async execution via
RabbitMQ. Add generation task runtime for lifecycle management, article
generation payload/queue abstractions, and template-assist queue publisher.
Refactor prompt generation service to support batch generation with
per-item error tracking. Update stream hub to bridge queue events to SSE.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

server/internal/tenant/app/template_service.go

@@ -13,6 +13,7 @@ import (
 	"github.com/geo-platform/tenant-api/internal/shared/auth"
 	"github.com/geo-platform/tenant-api/internal/shared/config"
 	"github.com/geo-platform/tenant-api/internal/shared/llm"
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
 	"github.com/geo-platform/tenant-api/internal/shared/response"
 	"github.com/geo-platform/tenant-api/internal/shared/stream"
 	"github.com/geo-platform/tenant-api/internal/tenant/repository"
@@ -22,13 +23,12 @@ type TemplateService struct {
 	pool            *pgxpool.Pool
 	templates       repository.TemplateRepository
 	llm             llm.Client
+	rabbitMQ        *rabbitmq.Client
 	knowledge       *KnowledgeService
 	streams         *stream.GenerationHub
 	streamEnabled   bool
 	articleTimeout  time.Duration
 	maxOutputTokens int64
-	jobs            chan generationJob
-	assistJobs      chan assistJob
 }
 
 type generationJob struct {
@@ -49,21 +49,12 @@ func NewTemplateService(
 	pool *pgxpool.Pool,
 	templates repository.TemplateRepository,
 	llmClient llm.Client,
+	rabbitMQClient *rabbitmq.Client,
 	knowledge *KnowledgeService,
 	streams *stream.GenerationHub,
 	cfg config.GenerationConfig,
 	llmMaxOutputTokens int64,
 ) *TemplateService {
-	queueSize := cfg.QueueSize
-	if queueSize <= 0 {
-		queueSize = 128
-	}
-
-	workerCount := cfg.WorkerConcurrency
-	if workerCount <= 0 {
-		workerCount = 1
-	}
-
 	articleTimeout := cfg.ArticleTimeout
 	if articleTimeout <= 0 {
 		articleTimeout = 8 * time.Minute
@@ -73,18 +64,12 @@ func NewTemplateService(
 		pool:            pool,
 		templates:       templates,
 		llm:             llmClient,
+		rabbitMQ:        rabbitMQClient,
 		knowledge:       knowledge,
 		streams:         streams,
 		streamEnabled:   cfg.StreamEnabled,
 		articleTimeout:  articleTimeout,
 		maxOutputTokens: llmMaxOutputTokens,
-		jobs:            make(chan generationJob, queueSize),
-		assistJobs:      make(chan assistJob, queueSize),
-	}
-
-	for i := 0; i < workerCount; i++ {
-		go svc.runGenerationWorker()
-		go svc.runAssistWorker()
 	}
 
 	return svc
@@ -272,7 +257,13 @@ func (s *TemplateService) Generate(ctx context.Context, templateID int64, req Ge
 		return nil, response.ErrForbidden(40301, "quota_insufficient", "generation quota insufficient, please upgrade your plan")
 	}
 
-	inputJSON, _ := json.Marshal(req.InputParams)
+	taskInputParams := cloneInputParams(req.InputParams)
+	taskInputParams["enable_web_search"] = req.EnableWebSearch
+	if req.WebSearchLimit != nil {
+		taskInputParams["web_search_limit"] = *req.WebSearchLimit
+	}
+	taskInputParams = attachTemplateSnapshot(taskInputParams, templateRecord)
+	inputJSON, _ := json.Marshal(taskInputParams)
 
 	tx, err := s.pool.Begin(ctx)
 	if err != nil {
@@ -369,26 +360,22 @@ func (s *TemplateService) Generate(ctx context.Context, templateID int64, req Ge
 	}
 
 	job := generationJob{
-		OperatorID:     actor.UserID,
-		TenantID:       actor.TenantID,
-		ArticleID:      articleID,
-		TaskID:         taskID,
-		ReservationID:  reservationID,
-		TemplateKey:    templateRecord.TemplateKey,
-		TemplateName:   templateRecord.TemplateName,
-		PromptTemplate: templateRecord.PromptTemplate,
-		Params:         cloneInputParams(req.InputParams),
-		InitialTitle:   initialTitle,
-		WebSearch: llm.WebSearchOptions{
-			Enabled: req.EnableWebSearch,
-		},
-	}
-	if req.WebSearchLimit != nil {
-		job.WebSearch.Limit = *req.WebSearchLimit
+		OperatorID:    actor.UserID,
+		TenantID:      actor.TenantID,
+		ArticleID:     articleID,
+		TaskID:        taskID,
+		ReservationID: reservationID,
+		Params:        taskInputParams,
+		InitialTitle:  initialTitle,
 	}
 
-	if err := s.enqueueGeneration(job); err != nil {
-		s.failGeneration(context.Background(), job, job.InitialTitle, "queue_enqueue", err)
+	if err := publishArticleGenerationTask(ctx, s.rabbitMQ, articleGenerationTaskEnvelope{
+		TaskID:    taskID,
+		TaskType:  "template",
+		TenantID:  actor.TenantID,
+		ArticleID: articleID,
+	}); err != nil {
+		s.failGeneration(context.Background(), job, initialTitle, "queue_enqueue", err)
 		return nil, response.ErrServiceUnavailable(50310, "generation_queue_unavailable", "generation queue is busy, please retry")
 	}
 
@@ -427,21 +414,6 @@ func updateGeneratedArticleState(
 	return nil
 }
 
-func (s *TemplateService) enqueueGeneration(job generationJob) error {
-	select {
-	case s.jobs <- job:
-		return nil
-	default:
-		return fmt.Errorf("generation queue is full")
-	}
-}
-
-func (s *TemplateService) runGenerationWorker() {
-	for job := range s.jobs {
-		s.executeGeneration(context.Background(), job)
-	}
-}
-
 func (s *TemplateService) executeGeneration(ctx context.Context, job generationJob) {
 	now := time.Now()
 	title := strings.TrimSpace(job.InitialTitle)
@@ -622,6 +594,8 @@ func formatGenerationFailure(stage string, genErr error) string {
 
 func generationFailureStageLabel(stage string) string {
 	switch stage {
+	case "task_load":
+		return "加载生成任务失败"
 	case "queue_enqueue":
 		return "任务入队失败"
 	case "llm_generate":