Status-less transport errors classified permanent in exactly-once subscriber retry

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

In `src/pubsub/src/subscriber/exactly_once_retry.rs`, error classification for exactly-once ack/nack/modack retries inspects `shared_err.status()`:

https://github.com/googleapis/google-cloud-rust/blob/f02b42c2ac5720334cd2b5335824bd4166264527/src/pubsub/src/subscriber/exactly_once_retry.rs#L37-L63

When a connection reset, broken pipe, or connection timeout occurs, `shared_err.status()` is `None`.

Because `shared_err.status()` is `None`, the match is bypassed and lines 52–63 convert all ack IDs into permanent `Err(AckError::Rpc)`. This contradicts `retry_policy.rs:105-114` where transport errors are intended to be retried.

Start in src/pubsub/src/subscriber/exactly_once_retry.rs at the error-classification logic, then compare it with retry_policy.rs:105-114. Verify how status-less connection resets, broken pipes, and timeouts are classified; done means these transport errors follow the retry policy instead of becoming permanent AckError::Rpc results.

Written by the indexing model from the issue text.