Tight infinite busy-loop in token cache refresh on short expiration tokens

Unconfirmed. This is the result of a search with Gemini, may be a false positive.

In `TokenCache::refresh_task`, the loop calculates sleep duration until expiry using `expiry.checked_duration_since(Instant::now())`:

https://github.com/googleapis/google-cloud-rust/blob/f02b42c2ac5720334cd2b5335824bd4166264527/src/auth/src/token_cache.rs#L180-L196

Specifically lines 186–194:

https://github.com/googleapis/google-cloud-rust/blob/f02b42c2ac5720334cd2b5335824bd4166264527/src/auth/src/token_cache.rs#L186-L194

```rust
if time_until_expiry > NORMAL_REFRESH_SLACK {
sleep(time_until_expiry - NORMAL_REFRESH_SLACK).await;
} else if time_until_expiry > SHORT_REFRESH_SLACK {
// If expiry is less than 4 mins, try to refresh every 10 seconds
// This is to handle cases where MDS **repeatedly** returns about to expire tokens.
sleep(SHORT_REFRESH_SLACK).await;
}
```

If the token provider returns a token expiring in `<= SHORT_REFRESH_SLACK` (10 seconds), or an already-expired token (where `checked_duration_since` returns `None`), neither `if` branch executes.

Because there is no fallback `else` branch or backoff sleep, the loop immediately ends its iteration and restarts, querying `token_provider.token().await` without any delay. If the metadata server (MDS) or auth backend repeatedly returns an about-to-expire token, this leads to an infinite 100% CPU busy loop that hammers the auth server and starves other async tasks.

Start in src/auth/src/token_cache.rs at TokenCache::refresh_task and inspect the expiry handling around lines 180–196. Exercise the loop with tokens expiring within 10 seconds or already expired, then verify that it no longer spins without a delay, repeatedly queries the token provider, or starves other async tasks.

Written by the indexing model from the issue text.