feat: Structured Output support

[kylexqian]

Added structured output support for chat. Tested all model providers + streaming.

We now also include the usage as part of the TextGenerationOutput. This was nice because it led to finding a bug in usage reporting for gemini in streaming.

Unit tests included as well

[kylexqian]

Based on 3rd party PR #169

[kylexqian]

Testing output

Structured output integration tests
Server : https://3.133.152.176
Schema : name (str) + age (int) + occupation (str)


-- OpenAI --------------------------------------------------
  [PASS]  OpenAI  | non-streaming | json_object  (prompt=24, completion=244)
         {
  "name": "Elena Marquez",
  "age": 32,
  "gender": "Female",
  "occupation": "Wildlife Biologist",
  "nationality": "Spanish",
  "residence": "Barcelona, Spain",
  "background": {
    "education": "PhD in Ecology from the University of Barcelona",
    "family": {
      "parents": ["Carlos Marquez", "Lucia Marquez"],
      "siblings": ["Miguel Marquez (younger brother)"]
    }
  },
  "personality_traits": [
    "curious",
    "adventurous",
    "empathetic",
    "meticulous"
  ],
  "hobbies": [
    "birdwatching",
    "rock climbing",
    "photography",
    "reading mystery novels"
  ],
  "notable_achievements": [
    "Discovered a new species of butterfly in the Pyrenees",
    "Published research on Mediterranean wetland restoration",
    "Recipient of the European Young Scientist Award (2022)"
  ],
  "fun_fact": "Elena speaks four languages: Spanish, Catalan, English, and French."
}
  [PASS]  OpenAI  | non-streaming | json_schema  (prompt=57, completion=17)
         {"name":"Lila Montgomery","age":32,"occupation":"Wildlife Photographer"}
  [PASS]  OpenAI  | streaming     | json_object  (prompt=24, completion=247)
         {
  "name": "Elena Marquez",
  "age": 32,
  "gender": "Female",
  "occupation": "Wildlife Biologist",
  "nationality": "Spanish",
  "background": {
    "birthplace": "Valencia, Spain",
    "education": "PhD in Ecology from the University of Barcelona",
    "languages": ["Spanish", "English", "French"]
  },
  "personality_traits": [
    "curious",
    "adventurous",
    "empathetic",
    "meticulous"
  ],
  "hobbies": [
    "birdwatching",
    "photography",
    "rock climbing",
    "reading mystery novels"
  ],
  "notable_achievements": [
    "Discovered a new species of butterfly in the Pyrenees",
    "Published research on migratory patterns of European storks",
    "Founded a non-profit for wildlife conservation"
  ],
  "current_residence": "Granada, Spain",
  "family": {
    "parents": ["Carlos Marquez", "Lucia Marquez"],
    "siblings": ["Miguel Marquez (older brother)"]
  }
}
  [PASS]  OpenAI  | streaming     | json_schema  (prompt=57, completion=17)
         {"name":"Lila Montgomery","age":32,"occupation":"Wildlife Photographer"}

-- Anthropic -----------------------------------------------
  [PASS]  Anthropic | non-streaming | json_schema  (prompt=208, completion=22)
         {"name": "Elena Vasquez", "age": 34, "occupation": "Marine Biologist"}
  [PASS]  Anthropic | streaming     | json_schema  (prompt=208, completion=22)
         {"name": "Elena Vasquez", "age": 34, "occupation": "Marine Biologist"}
  [PASS]  Anthropic -- json_object raises ValueError locally
         Anthropic models do not support response_format type 'json_object'. Use ResponseFormat(type='json_schema', json_schema={...}) with an explicit schema instead.

-- Gemini --------------------------------------------------
  [PASS]  Gemini  | non-streaming | json_object  (prompt=14, completion=138)
         {
  "person": {
    "name": "Anya Sharma",
    "age": 32,
    "occupation": "Quantum Physicist",
    "address": {
      "street": "789 Nebula Drive",
      "city": "Aethelburg",
      "zipCode": "90210",
      "country": "Eldoria"
    },
    "interests": [
      "Astrophysics",
      "Classical Music",
      "Hiking",
      "Ancient Languages"
    ],
    "isFictional": true
  }
}
  [PASS]  Gemini  | non-streaming | json_schema  (prompt=9, completion=14)
         {"name":"Alice","age":30,"occupation":"Engineer"}
  [PASS]  Gemini  | streaming     | json_object  (prompt=14, completion=136)
         {
  "person": {
    "name": "Anya Sharma",
    "age": 32,
    "occupation": "Quantum Physicist",
    "address": {
      "street": "789 Nebula Drive",
      "city": "Aethelburg",
      "zipCode": "90210",
      "country": "Eldoria"
    },
    "interests": [
      "Astrophysics",
      "Classical Music",
      "Hiking",
      "Ancient Languages"
    ],
    "isFictional": true
  }
}
  [PASS]  Gemini  | streaming     | json_schema  (prompt=9, completion=24)
         {
"name": "Alice", "age": 30, "occupation": "Software Engineer"
}

-- xAI / Grok ----------------------------------------------
  [PASS]  Grok    | non-streaming | json_object  (prompt=707, completion=448)
         {
  "name": "Elara Voss",
  "age": 34,
  "occupation": "Archaeologist",
  "background": "Born in a remote village, Elara discovered ancient ruins as a child, sparking her lifelong passion for uncovering lost civilizations. She now travels the world, deciphering forgotten languages and artifacts.",
  "personality": "Adventurous, intelligent, and fiercely independent, with a dry wit and a knack for getting into trouble.",
  "appearance": "Tall with sun-kissed skin, curly auburn hair, and piercing green eyes, often dressed in practical expedition gear."
}
  [PASS]  Grok    | non-streaming | json_schema  (prompt=774, completion=305)
          {"name":"Elara Voss","age":34,"occupation":"Quantum Physicist"}
  [PASS]  Grok    | streaming     | json_object  (prompt=707, completion=387)
         {"name":"Elara Voss","age":28,"occupation":"Adventurer and Archaeologist","background":"Born in a small coastal town, Elara discovered her passion for ancient mysteries after finding a hidden artifact in her grandmother's attic. She now travels the world, uncovering lost civilizations and dodging perilous traps.","personality":"Curious, brave, and quick-witted, with a sarcastic sense of humor that helps her in tight spots.","appearance":"Tall with athletic build, long auburn hair often tied back, piercing green eyes, and a scar on her left cheek from a close encounter with a booby-trapped tomb."}
  [PASS]  Grok    | streaming     | json_schema  (prompt=774, completion=222)
          {"name":"Alice Wonderland","age":28,"occupation":"Detective"}

------------------------------------------------------------
Results: 15/15 passed

Token usage:
  OpenAI  | non-streaming | json_object                prompt=   24  completion=  244
  OpenAI  | non-streaming | json_schema                prompt=   57  completion=   17
  OpenAI  | streaming     | json_object                prompt=   24  completion=  247
  OpenAI  | streaming     | json_schema                prompt=   57  completion=   17
  Anthropic | non-streaming | json_schema              prompt=  208  completion=   22
  Anthropic | streaming     | json_schema              prompt=  208  completion=   22
  Anthropic -- json_object raises ValueError locally   n/a
  Gemini  | non-streaming | json_object                prompt=   14  completion=  138
  Gemini  | non-streaming | json_schema                prompt=    9  completion=   14
  Gemini  | streaming     | json_object                prompt=   14  completion=  136
  Gemini  | streaming     | json_schema                prompt=    9  completion=   24
  Grok    | non-streaming | json_object                prompt=  707  completion=  448
  Grok    | non-streaming | json_schema                prompt=  774  completion=  305
  Grok    | streaming     | json_object                prompt=  707  completion=  387
  Grok    | streaming     | json_schema                prompt=  774  completion=  222

Not including this in unit tests atm -- just testing from a script