`thinking: disabled` leaves reasoning on for OpenAI models that reason by default

[CMGS]

An Anthropic-dialect thinking: {"type": "disabled"} sent to an OpenAI-family upstream maps to no effort at all (openai_engine::reasoning_effort documents that disabled leaves the vendor default). That was harmless while OpenAI models did not reason by default. It no longer is:

• gpt-6-sol, gpt-6-luna (and gpt-6-astra) reason when no effort is sent; live, temperature: 0.5 without an effort is a 400 on all three.
• Chat completions refuse function tools while a gpt-5.6 or gpt-6 model reasons: "Function tools with reasoning_effort are not supported for gpt-6-sol in /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to 'none'."

So POST /v1/messages {"model": "gpt-6-sol", "thinking": {"type": "disabled"}, "tools": [...]} against a chat-protocol gpt-6-sol gets a vendor 400. The same request against a Converse id (us.openai.gpt-6-sol) or a Responses-protocol model reasons and bills reasoning tokens the client disabled.

The chat surface's OpenRouter-style reasoning: {enabled: false} already maps to none, so the two ways of saying "no reasoning" behave differently today.

Proposal: map disabled to effort none on the OpenAI wires and Converse reasoning_config. openai_effort then clamps it per model: Astra becomes low, GPT-5.0 becomes minimal. This changes behavior for every OpenAI-family upstream, so it needs a decision before a PR.

[CMGS]

Fixed on main by #123 (squash d23c0e0); the follow-ups #124 (f16be65) and #126 (a30b31a) close the remaining edges.