Logging.has_run_logging returned early for every event type when
self.stream was True, so has_logged_async_failure / has_logged_sync_failure
were never set on streaming requests. Retried streamed calls re-enter
async_failure_handler once per attempt on the same logging object
(num_retries, router retries), and each attempt fired the failure
callbacks again, duplicating events to sinks like s3_v2 and Datadog.
Restrict the streaming early-return to the success event types, which
genuinely re-enter once per chunk, and keep the failure markers working
the same as non-streaming calls.
Resolves LIT-8542
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
25f257cc51