feat: stream chapter-name generation via SSE batching

Split chapter-name LLM requests into 100-chapter batches and stream
results back as SSE so large books (e.g. Shadow Slave: 2916 chapters)
never time out or truncate. Frontend shows live batch progress inline
and accumulates proposals as they arrive.

ui/src/routes/api/admin/text-gen/chapter-names/+server.ts

@@ -2,10 +2,11 @@
  * POST /api/admin/text-gen/chapter-names
  *
  * Admin-only proxy to the Go backend's chapter-name generation endpoint.
- * Returns AI-proposed chapter titles; does NOT persist anything.
+ * The backend streams SSE events; this handler pipes the stream through
+ * directly so the browser can consume it without buffering.
  */
 
-import { json, error } from '@sveltejs/kit';
+import { error } from '@sveltejs/kit';
 import type { RequestHandler } from './$types';
 import { log } from '$lib/server/logger';
 import { backendFetch } from '$lib/server/scraper';
@@ -28,6 +29,22 @@ export const POST: RequestHandler = async ({ request, locals }) => {
 		throw error(502, 'Could not reach backend');
 	}
 
-	const data = await res.json().catch(() => ({}));
-	return json(data, { status: res.status });
+	// Non-2xx: the backend returned a JSON error before switching to SSE.
+	if (!res.ok) {
+		const data = await res.json().catch(() => ({}));
+		return new Response(JSON.stringify(data), {
+			status: res.status,
+			headers: { 'Content-Type': 'application/json' }
+		});
+	}
+
+	// Pipe the SSE stream straight through — do not buffer.
+	return new Response(res.body, {
+		status: 200,
+		headers: {
+			'Content-Type': 'text/event-stream',
+			'Cache-Control': 'no-cache',
+			'X-Accel-Buffering': 'no'
+		}
+	});
 };