feat(ai-core): real vision image analysis

- llm: Adapter.Vision sends multimodal chat requests (text + image_url,
  data URLs supported) and logs the call as purpose 'vision'
- gateway /images/analyze now accepts image_url or image_base64 and
  returns real analysis instead of a placeholder message
- main wires the default LLM adapter as the vision analyzer
- httptest coverage for the vision request/response shape

backend/services/gateway/api/router.go

@@ -91,6 +91,11 @@ type AutomationEngine interface {
 	Trigger(ctx context.Context, id string) error
 }
 
+// VisionAnalyzer analyzes images with a multimodal model.
+type VisionAnalyzer interface {
+	AnalyzeImage(ctx context.Context, prompt, imageRef string) (string, error)
+}
+
 // ChatStorer persists chat sessions and messages.
 type ChatStorer interface {
 	CreateSession(ctx context.Context, id, userID, title string) (*store.Session, error)
@@ -117,6 +122,7 @@ type Dependencies struct {
 	Automation   *AutomationStore
 	AutomationEng AutomationEngine
 	Reminders    *ReminderStore
+	Vision       VisionAnalyzer
 	PluginList   []map[string]string
 }