feat: AI-generated shorts render pipeline

- Remotion composition (ShortsVideo) renders 1080x1920 9:16 shorts
  from script segments (hook/setup/conflict/insight/cta) with animated
  gradient background, word-by-word captions, progress bar
- render-worker.mjs: standalone poller (systemd hermyhq-render.service)
  TTS via edge-tts → Remotion render via Chrome headless → ffmpeg mux +
  thumbnail; ≤60s gate; marks YoutubeScript rendered
- API: POST /api/youtube/render (queue approved→render_queued),
  GET status, GET /render/file (stream video/thumb with Range support)
- Schema: YoutubeScript.videoUrl/thumbnailUrl + YoutubeRender model
- UI: youtube tab gains Renders view (preview/review/approve/reject/
  download), Render Shorts button on approved scripts
- Prisma db push applied to prod DB

prisma/schema.prisma

@@ -165,6 +165,10 @@ model YoutubeScript {
   seoChapters     Json?
   titleVariants   Json?
   thumbnailConcept String?
+  // Render output (shorts)
+  videoUrl        String?
+  thumbnailUrl    String?
+  render          YoutubeRender?
   createdAt       DateTime @default(now())
   updatedAt       DateTime @updatedAt
 
@@ -179,6 +183,22 @@ model YoutubeFeedback {
   date   DateTime @default(now())
 }
 
+model YoutubeRender {
+  id             String   @id @default(cuid())
+  scriptId       String   @unique
+  script         YoutubeScript @relation(fields: [scriptId], references: [id], onDelete: Cascade)
+  status         String   @default("queued") // queued | rendering | rendered | failed
+  videoPath      String?
+  thumbnailPath  String?
+  durationMs     Int?
+  error          String?
+  metadata       Json?
+  createdAt      DateTime @default(now())
+  updatedAt      DateTime @updatedAt
+
+  @@index([status])
+}
+
 model LongformScript {
   id               String   @id @default(cuid())
   title            String