Add Gemini reasoning support

[Davidyz]

Description

1. Properly add reasoning parameters in the requests.
2. handle reasoning output.
3. use feat(adapters): support extra fields in chat-completion response in OpenAI adapter #2359

Reference: https://ai.google.dev/gemini-api/docs/openai#thinking

Screenshots

Checklist

• I've read the contributing guidelines and have adhered to them in this PR
• I've added test coverage for this fix/feature
• I've run make all to ensure docs are generated, tests pass and my formatting is applied
• (optional) I've updated CodeCompanion.has in the init.lua file for my new feature
• (optional) I've updated the README and/or relevant docs pages

[Davidyz]

Since different models have different thinking budget, we might need to change validate so that it also accepts self in the parameters (something like validate(self, n) instead of the current validate(n). @olimorris would that be alright?

[Davidyz]

Btw, this may also be the culprit behind the Gemini adapter having occasional time out error: it defaults to use dynamic thinking, but the cc adapter didn't request for the reasoning summary and thought it timed out.

Maybe we should increase the default timeout for Gemini adapter or make it configurable?

[Davidyz]

The thinking_budget schema currently doesn't validate because the valid value depends on the model, and the current schema implementation doesn't pass self to the validate function.

[Davidyz]

A few things after the gemini-3-pro release:

1. Gemini-3-pro uses thinking_level ("low" or "high") instead of thinking_budget (integer). I've added the relevant schema, but haven't tested that on a live API.
2. They seem to have added support for the standard reasoning_effort field from the OpenAI API. It'll be mapped to the appropriate thinking_level or thinking_budget parameter on the server. This doesn't matter for the Gemini 3 model because it's also a string enum, but for the Gemini 2.5 series, we lose the fine control over the number of tokens used for reasoning.