fix(thinking): normalize disabled level-only models

[smartershining]

Summary

Normalize an explicit disabled-thinking request for pure level-only models before numeric budget clamping.

Some OpenAI-compatible models expose only reasoning levels and do not support disabling reasoning. A Claude request with thinking.type=disabled currently falls into the numeric budget path with a zero budget, which emits the misleading warning thinking: budget zero not allowed and relies on unsorted provider metadata for its fallback.

This change:

• keeps none unchanged when the model explicitly supports it;
• maps none to the canonical lowest supported level when a pure level-only model cannot disable thinking;
• preserves the existing budget fallback for hybrid models;
• avoids sending a level-only disabled sentinel through numeric budget clamping.

Tests

• go test -count=1 ./internal/thinking ./internal/thinking/provider/openai ./test
• go build -buildvcs=false -trimpath -o /tmp/cpa-pr1-build ./cmd/server

[github-actions]

This pull request targeted main.

The base branch has been automatically changed to dev.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 446727b405

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve provider-specific supported fallback levels

When a configured level-only model advertises only a provider-specific effort such as Levels: []string{"ultra"}, clampLevel(LevelMinimal, ...) ignores every unrecognized level and returns minimal, even though that value is not supported. Because this fallback is generated after level validation, the provider applier sends the invalid effort upstream; preserve an advertised fallback or validate the generated level before returning.

AGENTS.md reference: AGENTS.md:L25-L29

Useful? React with 👍 / 👎.