fix(ai): constrain GPT-6 Chat function calls to no reasoning

[opencode-agent]

Summary

• On native OpenAI Chat Completions, GPT-6 Sol/Luna requests containing function tools or tool history explicitly select reasoning_effort: "none" when no effort was chosen. Both models otherwise default to incompatible medium reasoning.
• Reject an explicitly configured higher Chat reasoning effort before dispatch instead of silently downgrading it; point callers to Responses for reasoning with tools.
• Leave Responses, tool-free Chat, and OpenAI-compatible provider routes on their existing options.

OpenAI contracts: GPT-6 Sol, GPT-6 Luna.

Verification

• bun test test/provider/openai-chat.test.ts test/provider/openai-responses.test.ts from packages/ai — 231 passed.
• bun typecheck from packages/ai and scoped oxlint — passed.
• bun run check at repo root: lint passed; repo-wide typecheck blocked by host Node.js 18.19.1, below @opencode/www Astro's >=22.12.0 requirement.

Requested by: @nexxeln (nexxel via Slack)