refactor: centralize sync polling in cron - #53

Why

#49 introduced one shared cron, but syncs still carried individual polling intervals and due times. The cron should determine when every enabled sync polls and retries.

How

Remove per-sync intervals, polling deadlines, source retry backoff and its failure counter, and obsolete schedule displays. Each cron round polls enabled syncs and drains their pages; successful and failed syncs wait for the next tick. Overlapping triggers share the round. Manual execution, pauses, checkpoints, lease fencing, capacity recovery, and queued delivery retry policies remain intact.

Preserve the released schema as the initial migration and remove obsolete columns through a numbered forward SQL migration. SQLite parses and executes the plain SQL migration statements on the application's existing connection; the runner records migration history and rolls back failed upgrades. createOpenSync automatically migrates an existing data directory. Upgrade coverage creates that directory using published 0.3.1, then verifies preserved checkpoints, poll history, queued deliveries and asset bytes, provider configuration, owner isolation, and the embedding host's database.

Host callers must remove intervalMs and references to nextDueAt; there is no replacement per-sync deadline. Stop the host and back up its data directory before upgrading. Reverting to the previous binary requires restoring that backup.