fix(provider): wrap default OpenAI screenshot selector

[sarath-menon]

Issue for this PR

Completes #169 after staging showed that OpenAI still sent inline screenshots.

Type of change

• Bug fix

What does this PR do?

Workers configure OpenAI with a run token and no provider key in the agent environment. That path selects languageModel(), bypassing the existing responses() wrapper. Wrap the default selector too, preserving the route, account isolation, cache, and cleanup.

How did you verify your code works?

The unchanged actual Provider.getLanguage regression fails on old code and passes with one upload, three file-reference requests without base64, preserved detail, and one deletion. All 22 image-reference tests and the package typecheck pass; Cubic review is clear.

How we tested in staging

We deployed the gateway first, then the released BrowserCode worker through the normal staging workflow; read-only AWS checks confirmed the matching image and READY runtime. The full existing API suite passed all 61 checks.

Two owned browser tasks used direct OpenAI gpt-6.1-sol and direct Anthropic claude-fable-5. Each took one screenshot, made three separate text-only tool turns, and returned the expected earlier image details. Each had four later model requests with one file reference and zero inline images. Datadog lifecycle logs record one upload and one successful deletion per task; test sessions, browsers, and workspaces were cleaned up.

OpenAI run: eab3ef01-5d4d-4245-9949-63699149fc82. Anthropic run: b57f33c0-8f24-4fa4-8daf-7246c2c6d302. These are shared-staging results, not production savings.

The released binary is 0.1.21-screenshot-files.2, deployed by browser-use/cloud#6269. No production deployment was performed.

Checklist

• I have tested my changes locally and in shared staging
• I have not included unrelated changes in this PR

[cubic-dev-ai]

All reported issues were addressed across 2 files

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>