[ML] Support Gemini thinking budget in inference API

[DonalEvans]

Adding support for configuring the thinkingBudget for Gemini 2.5 models when creating chat completion inference endpoints. The thinking_budget field is nested inside the thinking_config object in service_settings.

• Added ThinkingConfig class to contain the thinking_budget field. This results in a less flat structure for the PUT _inference/chat_completion/ call but will make adding support for include_thoughts easier in future
• Added extractOptionalInteger() method to ServiceUtils
• Unit tests for ThinkingConfig class
• Updated existing tests to account for the new object and field

These changes enable elastic/kibana#227590 to be completed

Specification PR: elastic/elasticsearch-specification#5257

Example usage:

PUT _inference/chat_completion/my_chat_completion_with_thinking_budget
{
  "service": "googlevertexai",
  "service_settings": {
    "service_account_json": <service account info>,
    "model_id": "gemini-2.5-pro",
    "location": "us-central1",
    "project_id": <project id>
  },
  "task_settings" : {
    "thinking_config": {
      "thinking_budget": 256
    }
  }
}

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[jonathan-buttner]

Great work! I left a few comments.

[jonathan-buttner]

Just wanted to capture our discussion earlier this week. After thinking about this more I think it probably makes sense to move the thinking budget settings to the task_settings instead of the service_settings. This gives us more flexibility in the future because the user could in theory override the task_settings on a per request basis. Because this is for chat completion which adheres to a strict schema format we won't be able to allow individual requests to override the task settings but they can at least set them during the inference endpoint creation request.

[jonathan-buttner]

Looking great, just a few comments.

[jonathan-buttner]

Great work!