### Summary

`DBSpendUpdateWriter.update_database()` awaits the spend-log insert but launches the user, key, team, organization, tag, agent, and daily spend updates with `asyncio.create_task()`. The returned coroutine completes before those updates run, and each helper failure is caught and logged at debug level inside the background task

### Current behavior

The writer contains:

~~~python asyncio.create_task( self._batch_database_updates(...) ) ~~~

`_batch_database_updates()` runs the entity and daily update helpers sequentially, but wraps every helper in its own `try/except Exception` and only emits a debug message. The task is not retained by the writer for a later await, retry, or graceful-shutdown drain

The cost callback therefore treats `await update_database()` as completion even though the durable entity counters may not have been queued or written

This is different from the spend-log write failure covered by [#33873](https://github.com/BerriAI/litellm/issues/33873), because the spend-log insert is awaited before this background task is created

### Reproduction

Patch one of the helper methods to fail and call `update_database()`:

~~~python writer._update_team_d…