fix: enforce English canonical text in AI-generated dish and recipe data

RecognizeDish() and buildRecipePrompt() were generating text in the
user's language and storing it in base tables, violating the project
rule that base tables always hold English canonical text.

- RecognizeDish(): hardcode English for dish_name; enrichDishInBackground()
  now correctly translates FROM English into all other languages
- buildRecipePrompt(): remove langName lookup, hardcode English for all
  text fields; drop unused locale import

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/internal/adapters/openai/recipe.go

@@ -7,7 +7,6 @@ import (
 	"strings"
 
 	"github.com/food-ai/backend/internal/adapters/ai"
-	"github.com/food-ai/backend/internal/infra/locale"
 )
 
 // goalNames maps internal goal codes to English descriptions used in the prompt.
@@ -57,18 +56,6 @@ func (c *Client) GenerateRecipes(ctx context.Context, req ai.RecipeRequest) ([]a
 }
 
 func buildRecipePrompt(req ai.RecipeRequest) string {
-	lang := req.Lang
-	if lang == "" {
-		lang = "en"
-	}
-	langName := "English"
-	for _, l := range locale.Languages {
-		if l.Code == lang {
-			langName = l.EnglishName
-			break
-		}
-	}
-
 	goal := goalNames[req.UserGoal]
 	if goal == "" {
 		goal = "weight maintenance"
@@ -100,7 +87,7 @@ func buildRecipePrompt(req ai.RecipeRequest) string {
 			strings.Join(req.AvailableProducts, "\n") + "\n"
 	}
 
-	return fmt.Sprintf(`You are a chef and nutritionist. Generate %d recipes in %s.
+	return fmt.Sprintf(`You are a chef and nutritionist. Generate %d recipes in English.
 
 User profile:
 - Goal: %s
@@ -114,7 +101,7 @@ Requirements for each recipe:
 - Include approximate macros per serving
 
 IMPORTANT:
-- All text fields (title, description, ingredient names, units, step descriptions, tags) MUST be in %s.
+- All text fields (title, description, ingredient names, units, step descriptions, tags) MUST be in English.
 - The "image_query" field MUST always be in English (it is used for stock-photo search).
 
 Return ONLY a valid JSON array without markdown or extra text:
@@ -131,7 +118,7 @@ Return ONLY a valid JSON array without markdown or extra text:
   "steps": [{"number": 1, "description": "...", "timer_seconds": null}],
   "tags": ["..."],
   "nutrition_per_serving": {"calories": 420, "protein_g": 48, "fat_g": 12, "carbs_g": 18}
-}]`, count, langName, goal, req.DailyCalories, restrictions, cuisines, productsSection, perMealCalories, langName)
+}]`, count, goal, req.DailyCalories, restrictions, cuisines, productsSection, perMealCalories)
 }
 
 func parseRecipesJSON(text string) ([]ai.Recipe, error) {

backend/internal/adapters/openai/recognition.go

@@ -111,11 +111,7 @@ Return ONLY valid JSON without markdown:
 // RecognizeDish uses the vision model to identify a dish and estimate its nutritional content.
 // Returns 3–5 ranked candidates so the user can correct mis-identifications.
 func (c *Client) RecognizeDish(ctx context.Context, imageBase64, mimeType, lang string) (*ai.DishResult, error) {
-	langName := langNames[lang]
+	prompt := `You are a dietitian and culinary expert.
-	if langName == "" {
-		langName = "English"
-	}
-	prompt := fmt.Sprintf(`You are a dietitian and culinary expert.
 
 Look at the dish photo and suggest 3 to 5 possible dishes it could be.
 Even if the first option is obvious, add 2–4 alternative dishes with lower confidence.
@@ -128,7 +124,7 @@ For each candidate specify:
 
 Sort candidates by descending confidence. First — most likely.
 
-Return dish_name values in %s.
+Return dish_name values in English.
 Return ONLY valid JSON without markdown:
 {
   "candidates": [
@@ -151,7 +147,7 @@ Return ONLY valid JSON without markdown:
       "confidence": 0.65
     }
   ]
-}`, langName)
+}`
 
 	text, err := c.generateVisionContent(ctx, prompt, imageBase64, mimeType)
 	if err != nil {