### Summary

A request can receive a successful provider response while the cost callback has no `response_cost`. In that branch the proxy releases the request's budget reservation to zero and does not write spend or increment spend counters

### Current behavior

In `litellm/proxy/hooks/proxy_track_cost_callback.py`, the callback writes spend only when `response_cost is not None`

When `response_cost is None`, it calls `_release_budget_reservation()` and then raises an internal cost-tracking exception for model calls. The outer exception handler logs the error and suppresses it. The database spend writer and spend-counter increment path are not called from this branch

An explicit numeric `response_cost=0` follows a different branch. The issue is specifically the missing-cost case, which can occur after a successful call when pricing or usage extraction did not produce a cost

### Reproduction

Use a request that creates a budget reservation, then invoke the cost callback with a successful logging object whose cost is missing

~~~python standard_logging_object = { "response_cost": None, "model": "configured-model", "status": "success", } ~~~

With mocks around the call…