feat: add image generation for OpenAI DALL-E and Gemini Imagen

New `/v1/images/generations` endpoint proxies DALL-E 2/3 (OpenAI)
and Imagen 3 (Gemini). Same auth/logging as chat completions.

- Add ImageGenerationRequest/Response models
- Extend Provider interface with ImageGeneration()
- OpenAI: forward to /v1/images/generations
- Gemini: call /v1beta/models/{model}:predict, map OpenAI params
- Circuit breaker wraps image gen like chat completions
- Model routing: dall-e* -> openai, imagen*/gemini* -> gemini
- Unsupported providers (deepseek/moonshot/grok/ollama) return error
- Fix pre-existing CachedContentTokenCount bug in StreamGemini

internal/providers/gemini.go

@@ -75,6 +75,116 @@ type GeminiFunctionResponse struct {
 	Response json.RawMessage `json:"response"`
 }
 
+func (p *GeminiProvider) ImageGeneration(ctx context.Context, req *models.ImageGenerationRequest) (*models.ImageGenerationResponse, error) {
+	// Gemini Imagen API: POST https://generativelanguage.googleapis.com/v1beta/models/{model}:predict
+	// Map OpenAI-style params to Gemini Imagen params
+
+	n := uint32(1)
+	if req.N != nil && *req.N > 0 {
+		n = *req.N
+	}
+
+	aspectRatio := "1:1"
+	if req.Size != nil {
+		aspectRatio = sizeToGeminiAspectRatio(*req.Size)
+	}
+
+	// Build Imagen request
+	imagenReq := map[string]interface{}{
+		"instances": []map[string]interface{}{
+			{"prompt": req.Prompt},
+		},
+		"parameters": map[string]interface{}{
+			"sampleCount": n,
+			"aspectRatio": aspectRatio,
+		},
+	}
+
+	// Model defaults to imagen-3.0-generate-001 if empty
+	model := req.Model
+	if model == "" {
+		model = "imagen-3.0-generate-001"
+	}
+
+	// Use v1beta for Imagen
+	baseURL := p.config.BaseURL
+	if !strings.Contains(baseURL, "v1beta") {
+		baseURL = strings.Replace(baseURL, "/v1", "/v1beta", 1)
+	}
+
+	url := fmt.Sprintf("%s/models/%s:predict?key=%s", baseURL, model, p.apiKey)
+
+	resp, err := p.client.R().
+		SetContext(ctx).
+		SetHeader("Content-Type", "application/json").
+		SetBody(imagenReq).
+		Post(url)
+
+	if err != nil {
+		return nil, fmt.Errorf("gemini imagen request failed: %w", err)
+	}
+
+	if !resp.IsSuccess() {
+		return nil, fmt.Errorf("Gemini Imagen API error (%d): %s", resp.StatusCode(), resp.String())
+	}
+
+	// Parse Imagen response
+	var imagenResp struct {
+		Predictions []struct {
+			MimeType           string `json:"mimeType"`
+			BytesBase64Encoded string `json:"bytesBase64Encoded"`
+		} `json:"predictions"`
+	}
+
+	if err := json.Unmarshal(resp.Body(), &imagenResp); err != nil {
+		return nil, fmt.Errorf("failed to parse Imagen response: %w", err)
+	}
+
+	respFormat := "url"
+	if req.ResponseFormat != nil && *req.ResponseFormat == "b64_json" {
+		respFormat = "b64_json"
+	}
+
+	var data []models.ImageData
+	for _, pred := range imagenResp.Predictions {
+		imgData := models.ImageData{}
+		if respFormat == "b64_json" {
+			imgData.B64JSON = pred.BytesBase64Encoded
+		} else {
+			// Build a data URI since Gemini returns base64, not a URL
+			mime := pred.MimeType
+			if mime == "" {
+				mime = "image/png"
+			}
+			imgData.URL = fmt.Sprintf("data:%s;base64,%s", mime, pred.BytesBase64Encoded)
+		}
+		data = append(data, imgData)
+	}
+
+	result := &models.ImageGenerationResponse{
+		Created: time.Now().Unix(),
+		Data:    data,
+	}
+
+	return result, nil
+}
+
+// sizeToGeminiAspectRatio converts OpenAI size format (e.g. "1024x1024") to Gemini aspect ratio (e.g. "1:1")
+func sizeToGeminiAspectRatio(size string) string {
+	switch size {
+	case "1024x1024":
+		return "1:1"
+	case "1024x1792":
+		return "9:16"
+	case "1792x1024":
+		return "16:9"
+	case "256x256", "512x512":
+		return "1:1"
+	default:
+		return "1:1"
+	}
+}
+
 func (p *GeminiProvider) ChatCompletion(ctx context.Context, req *models.UnifiedRequest) (*models.ChatCompletionResponse, error) {
 	// Gemini mapping
 	var contents []GeminiContent