Errors and cron check-ins from jobs never reach sentry

rq ends work horses with `os._exit`, which skips Sentry SDK's atexit flush; anything in the transport queue is dropped; now calls `sentry_sdk.flush()` before teardown.