Portals no longer stall before becoming ready

Before a portal starts serving, one of its instances has to volunteer to prepare the content types everything else depends on. The instances agree on who does the work by writing a small coordination record to the database.

That write could fail outright — not because of anything about the data, but because of how the write was described to PostgreSQL. Once the coordination record existed, every later attempt to update it was rejected, so nobody was ever chosen to do the work. The portal reported "preparing content types" and stayed there indefinitely: pages never loaded and the instance never became healthy.

The write is now described precisely enough that PostgreSQL accepts it in every case, so the coordination proceeds, the content types get prepared, and the portal comes up. The problem only affected records the platform writes for itself, which is why it surfaced as a startup stall rather than as an error anyone could see while using the portal.