Add GPT-6 Astra, Sol, and Luna model support

[flore2003]

What

Register OpenAI's GPT-6 chat models and apply the API constraints from the model guide:

• gpt-6-astra — flagship model. Reasoning efforts low, medium, high, and max. Chat Completions does not support tool calling.
• gpt-6-sol — coding and agentic work. Same effort range. Chat Completions function calling works only with reasoning disabled.
• gpt-6-luna — high-volume work. Same effort range and Chat Completions rule as Sol.

All three reason by default (reasoning.mode is always-on), so temperature and topP are rejected even when reasoning is omitted. minimal is clamped to low. Core max is sent as OpenAI max (earlier models still send xhigh). Dated snapshots such as gpt-6-sol-20260922 resolve to the base model.

On Chat Completions, Sol and Luna requests that include tools and omit reasoning send reasoning_effort: "none", which is the only effort that API accepts with function tools. Combining tools with reasoning throws and points at the Responses API.

Why

GPT-6 Astra, Sol, and Luna are the current flagship API models. Without registry entries, core-ai treated them as unknown models: standard effort clamping, max_tokens instead of max_completion_tokens, and no guard against Chat Completions requests the API rejects.

Testing

• npm run test -w @core-ai/openai
• npm run lint -w @core-ai/openai
• npm run check-types -w @core-ai/openai