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/cmd/cyrene/main.go

@@ -315,6 +315,7 @@ func main() {
 		gwDeps.ChatStore = chatStore
 	}
 	gwDeps.VoiceSvc = voiceSvc
+	gwDeps.Vision = visionAdapter{llmSelector.DefaultAdapter()}
 
 	wsHub := ws.NewHub(orch)
 	gwDeps.WSHub = wsHub
@@ -462,6 +463,17 @@ func (a *perfAdapter) History() []map[string]any {
 	return result
 }
 
+// visionAdapter adapts the LLM adapter to api.VisionAnalyzer.
+type visionAdapter struct{ llm *llm.Adapter }
+
+func (v visionAdapter) AnalyzeImage(ctx context.Context, prompt, imageRef string) (string, error) {
+	resp, err := v.llm.Vision(ctx, prompt, imageRef)
+	if err != nil {
+		return "", err
+	}
+	return resp.Content, nil
+}
+
 // automationSourceAdapter adapts api.AutomationStore to engine.RuleSource.
 type automationSourceAdapter struct{ s *api.AutomationStore }