fix: immediate AI loading feedback + auto-fallback for deprecated model

- BulkAiAnalyzer: set isAnalyzing=true immediately on click (before
  context fetch), so the spinner appears instantly instead of after
  1-2 seconds of silence
- analyze/route.ts: if configured model returns 400/404 (e.g., deprecated
  google/gemini-flash-1.5), automatically retry with google/gemini-2.0-flash-001
- batch-analyze/route.ts: same fallback logic

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

src/app/api/ai/analyze/route.ts

@@ -54,8 +54,22 @@ export async function POST(req: NextRequest) {
             }
         }
 
-        const client = new OpenRouterClient({ apiKey, model });
+        const FALLBACK_MODEL = "google/gemini-2.0-flash-001";
-        const result = await client.analyzeProduct(enrichedBody);
+        let client = new OpenRouterClient({ apiKey, model });
+        let result;
+        try {
+            result = await client.analyzeProduct(enrichedBody);
+        } catch (modelErr) {
+            // If configured model fails (e.g., deprecated), retry with fallback
+            const msg = modelErr instanceof Error ? modelErr.message : "";
+            if (model !== FALLBACK_MODEL && (msg.includes("400") || msg.includes("404") || msg.includes("OpenRouter error"))) {
+                console.warn(`[AI] Model "${model}" failed (${msg}), retrying with fallback ${FALLBACK_MODEL}`);
+                client = new OpenRouterClient({ apiKey, model: FALLBACK_MODEL });
+                result = await client.analyzeProduct(enrichedBody);
+            } else {
+                throw modelErr;
+            }
+        }
 
         return NextResponse.json(result);
     } catch (err) {

src/app/api/ai/batch-analyze/route.ts

@@ -107,14 +107,12 @@ REPONDS EN JSON VALIDE uniquement :
         const userPrompt = `${products.length} produits du rayon "${rayon}" (sur ${supplierStats?.totalProducts ?? products.length} au total).
 ${JSON.stringify(products, null, 2)}`;
 
-        const response = await fetch("https://openrouter.ai/api/v1/chat/completions", {
+        const FALLBACK_MODEL = "google/gemini-2.0-flash-001";
+        const tryModel = async (modelToTry: string) => fetch("https://openrouter.ai/api/v1/chat/completions", {
             method: "POST",
-            headers: {
+            headers: { "Authorization": `Bearer ${apiKey}`, "Content-Type": "application/json" },
-                "Authorization": `Bearer ${apiKey}`,
-                "Content-Type": "application/json",
-            },
             body: JSON.stringify({
-                model,
+                model: modelToTry,
                 messages: [
                     { role: "system", content: systemPrompt },
                     { role: "user", content: userPrompt }
@@ -124,6 +122,14 @@ ${JSON.stringify(products, null, 2)}`;
             }),
         });
 
+        let response = await tryModel(model);
+
+        // Fallback if configured model is deprecated/not found
+        if (!response.ok && response.status === 400 && model !== FALLBACK_MODEL) {
+            console.warn(`[batch-analyze] Model "${model}" returned 400, retrying with fallback ${FALLBACK_MODEL}`);
+            response = await tryModel(FALLBACK_MODEL);
+        }
+
         if (response.status === 429) {
             const retryAfter = response.headers.get("Retry-After") || response.headers.get("x-ratelimit-reset-requests");
             const waitSeconds = retryAfter ? parseInt(retryAfter, 10) : 60;

src/features/grid/components/bulk-ai-analyzer.tsx

@@ -41,6 +41,10 @@ export function BulkAiAnalyzer() {
         const { rows } = useGridStore.getState();
         isCancelledRef.current = false;
 
+        // Show immediate visual feedback before any async work
+        setIsAnalyzing(true);
+        setProgress({ current: 0, total: 0, message: "Chargement du contexte...", errors: 0 });
+
         // 1. Fetch AI Context for the Supplier
         let supplierContext = "";
         if (supplierCode) {
@@ -166,7 +170,6 @@ export function BulkAiAnalyzer() {
 
         // 2c. Context profiling en micro-batches asynchrones
         //    Le score est déjà calculé par score-engine.ts (champ row.score sur chaque ProductRow).
-        setIsAnalyzing(true);
         setProgress({ current: 0, total: payloadsWithSales.length, message: "Calcul du contexte...", errors: 0 });
 
         const SCORING_BATCH_SIZE = 25;