Make collage selection automatic and restore dashboard navigation order

docs/AUTOMATIC_COLLAGE.md

@@ -0,0 +1,55 @@
+# Automatic collage
+
+The navigation starts with Dashboard, followed by Gallery, Photo Sources, Upload,
+Templates, Layout Studio, Automation, Playlists, History, Connections and Settings.
+
+## Web workflow / 界面操作
+
+在「拼图」选择照片来源与数量（1–12），填写可选标题或排版要求，点击
+「随机选图并生成预览」。默认使用启用的数据源；没有数据源时使用本地上传照片。
+AI 只接收图片尺寸、内部资产 ID 和排版文字，不发送照片。
+AI 未配置或请求失败会明确报错，不会偷偷切换引擎；取消 AI 勾选可用自动网格。
+预览不会上传或切换画框，确认效果后再点「显示到画框」。
+
+Choose a source and count, then create a preview. Only dimensions, asset IDs and
+layout instructions are sent to AI. AI configuration/errors are explicit; uncheck
+AI for deterministic grid layout. Preview never changes the frame. Canvas
+orientation follows the target device. Manual layout is under Advanced.
+
+「创建自动化任务」保存来源和数量，而不是固定图片列表。任务初始暂停，
+请在自动化中编辑执行时间后启用。每次运行重新随机选图，但不同运行之间可能重复；
+单次不重复选同一照片。照片不足时失败，不重复图片凑数。
+NAS 来源继续使用人物、方向与质量筛选；一次最多检查数量的五倍候选图片。
+输入图缓存与生成图在 NAS；原始 NAS 照片不会被修改或删除。
+定时拼图沿用云库每任务保留最近一张成功图的策略，下一张确认成功前不删除上一张。
+
+Create automation stores the source/count, initially paused. Edit its schedule
+and enable it. Each execution selects new candidates; repeats across executions
+are possible. Too few usable photos fails before upload. Existing unknown-write,
+restoration and cloud-retention safeguards apply.
+
+## Public API
+
+`POST /api/v1/render` accepts this additive content type, returning HTTP 202 and
+an operation ID. `POST /api/v1/display` uses the same content inside its existing
+request envelope. Scheduler `source` accepts the same object; job previews are
+read-only.
+
+```json
+{
+  "type": "auto_collage",
+  "source_id": "nas-photos",
+  "count": 4,
+  "title": "Our moments",
+  "orientation": "auto",
+  "style": "grid",
+  "use_ai": true,
+  "prompt": "Feature one photo with smaller supporting photos",
+  "profile": "color"
+}
+```
+
+`source_id: null` chooses local uploaded assets (generated and cache images are
+excluded). AI configuration is shared with existing Layout Studio. No new API key
+or vendor dependency is introduced. The OpenAPI contract test records the new
+content schema while keeping all existing schema fields unchanged.

docs/UI_LAYOUT_AUDIT.md

@@ -1,7 +1,7 @@
 # UI layout audit
 
 The Mock-only `scripts/ui-layout-audit.cjs` checks all 11 navigation pages,
-local/cloud gallery layouts and six modal states at 1440, 768 and 390 CSS pixels,
+local/cloud gallery layouts and six modal states plus expanded manual collage at 1440, 768 and 390 CSS pixels,
 in Chinese/English and light/dark themes. It checks page horizontal overflow and
 at least 8 pixels between visible input/select/textarea/button rectangles, and
 44-pixel control heights on mobile.
@@ -32,5 +32,5 @@ node scripts/ui-layout-audit.cjs
 
 Set `NODE_PATH` to the installed Playwright runtime if needed. Optional
 `FRAME_FLOW_SCREENSHOT_DIR` saves fixture-only screenshots for visual review.
-The full audit passed 217 checks with no console errors. This complements
+The full audit passed 229 checks with no console errors. This complements
 `ui-smoke.cjs` and `ui-integrations.cjs`; it does not replace functional tests.