Startup no longer stalls on large meshes

The permission self-heal that runs at every startup used to sweep all partitions in one long operation. On large installations that single operation could outlast the database timeout, leaving the migration stuck in a restart loop while smaller installations sailed through.

The sweep now runs as a series of small batches, each finishing quickly and releasing its lock before the next begins. Startup time no longer grows into a timeout as your mesh grows, and several servers booting at once no longer queue behind one long sweep.