Flush buffer in streaming interface before writing zip data

We ran into a really obscure issue when working on ansible/receptor#683.

I'll try to make this at least somewhat digestable.

Due to a bug in Kubernetes, AWX can't currently run jobs longer than 4 hours when deployed into Kubernetes. More context on that in ansible/awx#11805

To address this issue, we needed a way to restart from a certain point in the logs. The only mechanism Kubernetes provides to do this is by passing "sinceTime" to the API endpoint for retrieving logs from a pod.

Our patch in ansible/receptor#683 worked when we ran it locally, but in OpenShift, jobs errored when unpacking the zip stream at the end of the results of "ansible-runner worker". Upon further investigation this was because the timestamps of the last 2 lines were exactly the same:

```
2022-11-09T00:07:46.851687621Z {"status": "successful", "runner_ident": "1"}
2022-11-08T23:07:58.648753832Z {"zipfile": 1330}
2022-11-08T23:07:58.648753832Z UEsDBBQAAAAIAPy4aFVGnUFkqQMAAIwK....
```

After squinting at this code for a bit I noticed that we weren't flushing the buffer here like we do in the event_handler and other callbacks that are fired in streaming.py. The end. Ugh.

ansible_runner/utils/streaming.py

@@ -51,6 +51,7 @@ def stream_dir(source_directory, stream):
             else:
                 target = stream
             target.write(json.dumps({"zipfile": zip_size}).encode("utf-8") + b"\n")
+            target.flush()
             with Base64IO(target) as encoded_target:
                 for line in source:
                     encoded_target.write(line)