Add error handling for unknown fields for chat completion detection request

[swith004]

What does your PR do?

This PR adds a container attribute from the serde framework to handle unknown fields passed into to a request for the chat completion detection (including the message and the detection section). Unit tests are also added asserting the case used for the scenario used in the git issue.

How this was tested

In addition to new validation unit tests, the request was modified to have an additional field in parts of the request, and was compared with an current orchestrator version deployed on openshift, versus the locally updated version.

Example:

curl -X 'POST' \
>   'http://localhost:8033/api/v2/chat/completions-detection' \
>   -H 'accept: application/json' \
>   -H 'Content-Type: application/json' \
>   -d '{
>   "messages": [
>     {
>       "content": "this is a nice sentence",
>       "role": "user",
>       "name": "string",
>       "additional": "test"
>     }
>   ],
>   "model": "/llama_3_storage/hf/8b_instruction_tuned",
>   "n": 1,
>   "temperature": 1,
>   "top_p": 1,
>   "user": "user-1234",
>   "detectors": {
>     "input": {
>       "en_syntax_slate.38m.hap": {},
>       "en_syntax_rbr_pii": {}
>     }
>   }
> }'
{"code":422,"details":"messages[0].additional: unknown field `additional`, expected one of `role`, `content`, `name`, `refusal`, `tool_calls`, `tool_call_id` at line 7 column 18"}

curl -X 'POST' \
>   'http://localhost:8033/api/v2/chat/completions-detection' \
>   -H 'accept: application/json' \
>   -H 'Content-Type: application/json' \
>   -d '{
>   "messages": [
>     {
>       "content": "this is a nice sentence",
>       "role": "user",
>       "name": "string"
>     }
>   ],
>   "model": "/llama_3_storage/hf/8b_instruction_tuned",
>   "additional_field": "test",
>   "n": 1,
>   "temperature": 1,
>   "top_p": 1,
>   "user": "user-1234",
>   "detectors": {
>     "input": {
>       "en_syntax_slate.38m.hap": {},
>       "en_syntax_rbr_pii": {}
>     }
>   }
> }'
{"code":422,"details":"additional_field: unknown field `additional_field`, expected one of `messages`, `model`, `store`, `metadata`, `frequency_penalty`, `logit_bias`, `logprobs`, `top_logprobs`, `max_tokens`, `max_completion_tokens`, `n`, `presence_penalty`, `response_format`, `seed`, `service_tier`, `stop`, `stream`, `stream_options`, `temperature`, `top_p`, `tools`, `tool_choice`, `parallel_tool_calls`, `user`, `best_of`, `use_beam_search`, `top_k`, `min_p`, `repetition_penalty`, `length_penalty`, `early_stopping`, `ignore_eos`, `min_tokens`, `stop_token_ids`, `skip_special_tokens`, `spaces_between_special_tokens`, `detectors` at line 10 column 20"}

curl -X 'POST' \
>   'http://localhost:8033/api/v2/chat/completions-detection' \
>   -H 'accept: application/json' \
>   -H 'Content-Type: application/json' \
>   -d '{
>   "messages": [
>     {
>       "content": "this is a nice sentence",
>       "role": "user",
>       "name": "string"
>     }
>   ],
>   "model": "/llama_3_storage/hf/8b_instruction_tuned",
>   "n": 1,
>   "temperature": 1,
>   "top_p": 1,
>   "user": "user-1234",
>   "detectors": {
>     "input": {
>       "en_syntax_slate.38m.hap": {},
>       "en_syntax_rbr_pii": {}
>     },
>    "additional_detector_field": "test"
>   }
> }'
{"code":422,"details":"detectors.additional_detector_field: unknown field `additional_detector_field`, expected `input` or `output` at line 19 column 30"}

[declark1]

LGTM. I noticed that the new and existing tests in chat_completions_detection.rs are async fn but actually don't need to be async. Can you drop the async and change #[tokio::test] attribute to #[test]?

[swith004]

LGTM. I noticed that the new and existing tests in chat_completions_detection.rs are async fn but actually don't need to be async. Can you drop the async and change #[tokio::test] attribute to #[test]?

Will do! - thanks for the review!

[evaline-ju]

LGTM