fix: text-gen chapter-names truncation and bad title format

- Default max_tokens to 4096 for chapter-names so large chapter lists
  are not cut off mid-JSON by the model's token limit
- Rewrite system prompt to clarify placeholder semantics ({n} = number,
  {scene} = scene hint) and explicitly forbid echoing the number inside
  the title field — prevents "Chapter 1 - 1: ..." style duplications
- UI: surface raw_response in the error area when chapters:[] is returned
  so the admin can see what the model actually produced

ui/src/routes/admin/text-gen/+page.svelte

@@ -71,6 +71,11 @@
 			);
 			chRawResponse = body.raw_response ?? '';
 			chUsedModel = body.model ?? '';
+			// If backend returned chapters:[] but we have a raw response, the model
+			// output was unparseable (likely truncated). Treat it as an error.
+			if (chProposals.length === 0 && chRawResponse.trim().length > 0) {
+				chError = 'Model response could not be parsed (output may be truncated). Raw response shown below.';
+			}
 		} catch {
 			chError = 'Network error.';
 		} finally {
@@ -287,6 +292,9 @@
 
 				{#if chError}
 					<p class="text-sm text-(--color-danger) bg-(--color-danger)/10 rounded-lg px-3 py-2">{chError}</p>
+					{#if chRawResponse}
+						<pre class="text-xs bg-(--color-surface-2) border border-(--color-border) rounded-lg p-3 overflow-auto max-h-48 text-(--color-muted) whitespace-pre-wrap break-words">{chRawResponse}</pre>
+					{/if}
 				{/if}
 			</div>