Duplicated data with same publish_ts but different message id!

**Setup**
- Pulsar 2.7.0, 3-node cluster
- Pulsar websocket api embedded within broker

**Describe the bug**
Client device is using websocket to send event-trigger data to pulsar. Network transmission from device to pulsar is in excess of 100ms on average. There is a retry mechanism if fail to send data. This is the snippet:

```python
...
self.wsDict[fulltopic] = websocket.create_connection(fulltopic)
jsd = json.dumps({
'payload' : base64.b64encode(json.dumps(data['Data'], ensure_ascii=False).encode('utf-8')).decode('utf-8'),
'context' : 5
})
self.wsDict[fulltopic].send(jsd)
res = self.wsDict[fulltopic].recv()
if self.wsDict[fulltopic].connected:
response = json.loads(res)
if response['result'] == 'ok':
print('ok')
else:
# retry mechanism
else:
# retry mechanism
...
```

Checking back pulsar table, I found sometimes there are **duplicated data of the same publish timestamp but different message id**. Below are the relevant rows, in format (message id, publish timestamp, data):
```
...
(37970857,97,-1,-1) :: 1658156520335 :: b'{"EventStartUTCTime": "2022-07-18 15:01:53", "EventEndLocalTime": "2022-07-18 17:01:53", "EventStartTime": 1658156513, "PeopleId": 1, "EventEndUTCTime": "2022-07-18 15:01:53", "UploadedLocalDateTime": "2022-07-18 17:01:53", "EventEndTime": 1658156513, "EventStartLocalTime": "2022-07-18 17:01:53", "UploadedUTCDateTime": "2022-07-18 15:01:53", "PeopleTypeId": 1}'
...
(37970857,149,-1,-1) :: 1658156520335 :: b'{"EventStartUTCTime": "2022-07-18 15:01:53", "EventEndLocalTime": "2022-07-18 17:01:53", "EventStartTime": 1658156513, "PeopleId": 1, "EventEndUTCTime": "2022-07-18 15:01:53", "UploadedLocalDateTime": "2022-07-18 17:01:53", "EventEndTime": 1658156513, "EventStartLocalTime": "2022-07-18 17:01:53", "UploadedUTCDateTime": "2022-07-18 15:01:53", "PeopleTypeId": 1}'
...
```

I rule out the possibility of client device executing the retry mechanism because it is physically not possible to send the data 2 times at the exact timestamp up to the milliseconds.

This issue happens randomly on my other client devices as well, sometimes up to 3 times duplicates.

I have not set [`message deduplication`](https://pulsar.apache.org/docs/cookbooks-deduplication/) on the pulsar server. I do not think it will help because [wiki](https://github.com/apache/pulsar/wiki/PIP-6:-Guaranteed-Message-Deduplication#design) says deduplication is based on message `sequenceId`, which are clearly different above! Plus, deduplication seems to be not supported on pulsar websocket api (correct me if I am wrong).

**To Reproduce**
Unfortunately, I do not know a reliable way to reproduce the issue. I have tried simulate a device spamming messages into pulsar at quick succession. It happened only 2 times out of 20,000 sends, rather randomly.

**Expected behavior**
Duplicate should not happen.

What can I do to solve this / diagnose further?

The report identifies the Pulsar WebSocket API, a Python client, message IDs, publish timestamps, and message deduplication documentation, but no repository file or reliable reproduction. Start by reviewing the WebSocket API and deduplication documentation, then investigate how retries could produce the differing message IDs. Done means identifying a reproducible cause and documenting a diagnosis or fix.

Written by the indexing model from the issue text.