Assistant: Support defaultThinkingEffort in custom model definitions (providers.json) #116

Problem Description:

When configuring custom language models in ~/.posit/ai/providers.json, users can declare available thinking levels via thinkingEffortLevels, for example:

"thinkingEffortLevels": [
  "low",
  "medium",
  "high",
  "max"
]

However, there is currently no way to declare a default thinking effort for a model (e.g. "defaultThinkingEffort": "high").

Impact:

1. Initial conversation state defaults to null: When a new conversation begins, the session state initializes thinkingEffort: null.
2. Missing reasoning payload: Unless the user manually clicks the thinking effort selector in the UI for every new conversation, the outgoing request to the model provider omits the reasoning parameter entirely (reasoningEffort is not included in the payload).
3. Suboptimal performance for reasoning models: Controllable reasoning models (such as o-series, GPT-5/6, Gemini 2.5/3.x, Qwen, etc.) fall back to zero or minimal reasoning effort by default, defeating the purpose of configuring a reasoning model.

Proposed Solution:

Add support for an optional defaultThinkingEffort property in model definitions within providers.json:

{
  "id": "my-provider/my-reasoning-model",
  "name": "My Reasoning Model",
  "protocol": "openai-responses",
  "thinkingEffortLevels": ["low", "medium", "high", "max"],
  "defaultThinkingEffort": "high"
}

When a conversation selects this model, Positron Assistant should automatically initialize the conversation's thinkingEffort with the declared defaultThinkingEffort instead of null.