Synchronize completion cache updates with ALPM handle lifetime

[Copilot]

yay -Syu could crash after a successful transaction when the completion cache refresh was still traversing sync databases while the main flow released the shared ALPM handle. The race was caused by a fire-and-forget goroutine in OperationService.Run with no synchronization before cleanup.

• Serialize completion cache teardown with command teardown

  • Extract the completion refresh launch into a small helper that returns a wait function.
  • Start the refresh asynchronously as before, but defer the wait in OperationService.Run so the method does not return until the cache update is done.
  • This keeps the existing background behavior while ensuring main() cannot reach dbExecutor.Cleanup() early.

• Add focused coverage for the synchronization point

  • Add unit tests for the helper to verify:
    • no background work starts when the cache is fresh
    • the returned wait function blocks until the background refresh completes

• Representative change

  waitForCompletionUpdate := o.startCompletionUpdate(ctx, run)
  if waitForCompletionUpdate != nil {
      defer waitForCompletionUpdate()
  }