feat: allow json_schema in response format and add test

[drbh]

This PR improve the response_format.type = "json_schema" to allow the schema to be specified with the key value or json_schema, prior to these changes only the value key would be allowed.

These changes enable the two following request to succeed:

import requests
import json

# API endpoint
url = "http://localhost:3000/v1/chat/completions"

# Your payload
payload = {
   "temperature": 0,
   "response_format": {
       "value": {
           "name": "test",
           "strict": True,
           "schema": {
               "type": "object",
               "properties": {"status": {"type": "string"}},
               "required": ["status"],
               "additionalProperties": False,
           },
       },
       "type": "json_schema",
   },
   "messages": [
       {
           "role": "system",
           "content": "You are a helpful assistant. You answer with a JSON output with a status string containing the value \u0027OK\u0027",
       },
       {"role": "user", "content": "Please tell me OK"},
   ],
   "model": "tgi",
   "max_completion_tokens": 8192,
}

payload1 = {
  "temperature": 0,
  "response_format": {
      "json_schema": {
          "name": "test",
          "strict": True,
          "schema": {
              "type": "object",
              "properties": {"status": {"type": "string"}},
              "required": ["status"],
              "additionalProperties": False,
          },
      },
      "type": "json_schema",
  },
  "messages": [
      {
          "role": "system",
          "content": "You are a helpful assistant. You answer with a JSON output with a status string containing the value \u0027OK\u0027",
      },
      {"role": "user", "content": "Please tell me OK"},
  ],
  "model": "tgi",
  "max_completion_tokens": 8192,
}

# Send request
for payload in [payload, payload1]:
  response = requests.post(url, json=payload)

  # Print response
  print("Status Code:", response.status_code)
  print("Response JSON:", json.dumps(response.json(), indent=4))

[Narsil]

Can you find a way to remove ALL serde(untagged), and have simpler hierarchy ?

I feel like unit tests with valid schemas and how we expect to parse them should exist in here.

[Narsil]

This seems like we just want serde(alias="value") if I read correctly.

[Narsil]

This is extremely fishy. untagged + recursive descent seems like a great way to shoot ourselves.