Model catalog reaches the database again

Deployments configure which AI content partitions are served from the database (Features:StaticRepoSync:Partitions), and the model/provider catalog is one of them: serving it from the database is what makes it durable, queryable, and manageable through git-governed sync.

On deployed portals that configuration was silently ignored. The AI engine activates as a module at boot, and modules read the deployment's configuration through the mesh builder — but the portal's boot path never handed its configuration over. The engine saw "nothing configured", fell back to in-memory serving (the documented answer for an absent key), and skipped registering its content sources. The result: sixteen models serving happily from memory, zero rows in the model schema, no provider schema at all, and nothing in any log naming why — every part behaved correctly for the inputs it saw.

What changed

The portal hands its configuration to the module system before any module installs, so module activations now see exactly what the deployment configured. A regression test pins the hand-over.

What you will notice

The Provider and Model partitions materialize in the database as configured, which unblocks git-governed model catalog management (Provider/_GitSync) and makes provider/model changes durable across restarts instead of re-projected from configuration each boot.