fix: serialise local Cog model calls with a queue to prevent 409 conflicts

utils/replicate-translate.ts

@@ -90,12 +90,13 @@ async function translateViaCloud(
     return extractText(data.output as ReplicateOutput);
 }
 
-async function translateViaLocal(
-    text: string,
-    targetLanguage: string
-): Promise<string> {
-    const settings = readSettings();
+// The local Cog model handles exactly one prediction at a time.
+// This queue ensures calls are strictly serialised regardless of
+// how many batches / concurrent requests come in.
+let localQueue: Promise<void> = Promise.resolve();
 
+async function callLocalModel(text: string, targetLanguage: string): Promise<string> {
+    const settings = readSettings();
     if (!settings.jigsawApiKey) throw new Error("JigsawStack API key not configured");
 
     const endpoint = (settings.localEndpoint || "http://localhost:5030/predictions").replace(/\/$/, "");
@@ -118,14 +119,24 @@ async function translateViaLocal(
     }
 
     const data: CogOutput = await response.json();
-
     if (data.error) throw new Error(`Local model error: ${data.error}`);
 
-    // Cog wraps output in { status, output } or returns output directly
     const output = data.output !== undefined ? data.output : data;
     return extractText(output);
 }
 
+async function translateViaLocal(
+    text: string,
+    targetLanguage: string
+): Promise<string> {
+    // Enqueue — wait for any in-flight local call to finish first
+    let resolve!: () => void;
+    const slot = new Promise<void>(r => { resolve = r; });
+    const turn = localQueue.then(() => callLocalModel(text, targetLanguage));
+    localQueue = turn.then(resolve, resolve); // advance queue whether success or error
+    return turn;
+}
+
 export async function replicateTranslate(
     text: string,
     targetLanguage: string