fix(sw): stop silently dropping pending events when no deliveryUrl

[joelshejar]

Summary

Fixes a silent data-loss bug in the Service Worker handoff path.

Two compounding bugs:

1. `ServiceWorkerClient.register()` never forwarded `deliveryUrl` to the SW config message, so even when apps configured one, the SW couldn't see it.
2. When `config.deliveryUrl` was unset (the only path the SW ever saw), the SW would mark every pending event as 'delivered' without sending it anywhere — then the cleanup transaction deleted them from IndexedDB. The inline comment claiming "events will be picked up by the next Hub" was wrong: by the time another Hub started, the SW had already deleted them.

What changes

• Add `deliveryUrl` to `ServiceWorkerConfig`.
• `ServiceWorkerClient.register()` now forwards `deliveryUrl` to the SW.
• SW drain handler only marks entries delivered when `fetch()` returns ok. Without a `deliveryUrl`, leave pending entries in the outbox so the next Hub session drains them. TTL-expired entries are still GC'd.

Test plan

• `pnpm --filter @tabmesh/core exec vitest --run` — 99/99 pass (was 98/98)
• New test: `forwards deliveryUrl to the service worker config`
• Existing config-shape test updated to expect the new field
• Biome clean

Follow-up

• The full SW handoff (Background Sync firing → IDB drain → fetch → next Hub picks up) should be exercised in the upcoming Playwright harness.

🤖 Generated with Claude Code