HTTP Errors are not Correctly Exposed by the Python Client SDK

[gzxu]

When using the tritonclient.http or tritonclient.http.aio package to call a Triton server behind a gateway, and the gateway returns a non-200 status code with a message, the message might not be correctly raised to callers.

As shown in https://github.com/triton-inference-server/client/blob/main/src/python/library/tritonclient/http/_utils.py#L35C1-L59C20, the library assumes the response body to be a JSON with an error field, but it's not always the case when the Triton server is behind a gateway.

    if response.status_code != 200:
        body = None
        try:
            body = response.read().decode("utf-8")
            error_response = (
                json.loads(body)
                if len(body)
                else {"error": "client received an empty response from the server."}
            )
            return InferenceServerException(
                msg=error_response["error"], status=str(response.status_code)
            )
        except Exception as e:
            return InferenceServerException(
                msg=f"an exception occurred in the client while decoding the response: {e}",
                status=str(response.status_code),
                debug_details=body,
            )

I would suggest including the whole response body even if it cannot be decoded as a JSON

[gzxu]

Thanks for fixing!