Fix blog startup deadlock: use direct DB instead of self-HTTP call

ghost_sync was calling blog's own /internal/data/page-config-ensure via
HTTP during startup, but the server isn't listening yet — causing a retry
loop that times out Hypercorn. Replace with direct DB insert using the
existing session.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

blog/bp/blog/ghost/ghost_sync.py

@@ -248,13 +248,24 @@ async def _upsert_post(sess: AsyncSession, gp: Dict[str, Any], author_map: Dict[
     finally:
         sess.autoflush = old_autoflush
 
-    # Auto-create PageConfig for pages (blog now owns this table)
+    # Auto-create PageConfig for pages (blog owns this table — direct DB,
+    # not via HTTP, since this may run during startup before the server is ready)
     if obj.is_page:
-        await fetch_data(
-            "blog", "page-config-ensure",
-            params={"container_type": "page", "container_id": obj.id},
-            required=False,
-        )
+        from sqlalchemy import select
+        from shared.models.page_config import PageConfig
+        existing = (await sess.execute(
+            select(PageConfig).where(
+                PageConfig.container_type == "page",
+                PageConfig.container_id == obj.id,
+            )
+        )).scalar_one_or_none()
+        if existing is None:
+            sess.add(PageConfig(
+                container_type="page",
+                container_id=obj.id,
+                features={},
+            ))
+            await sess.flush()
 
     return obj, old_status