fix(js/plugins/compat-oai): omit visualDetailLevel from OpenAI request bodies

[cpruijsen]

toOpenAIRequestBody() now omits visualDetailLevel in the existing config destructure, the same way apiKey, maxOutputTokens, topK, and version are already omitted. Image detail mapping is unchanged.

config.visualDetailLevel is applied to image_url.detail in toOpenAIMessages(), but toOpenAIRequestBody() did not pull it out of request.config before spreading restOfConfig into the Chat Completions body. OpenAI then returned 400 Unknown parameter: 'visualDetailLevel'.

Fixes #4592.

Checklist

• PR title follows https://www.conventionalcommits.org/en/v1.0.0/
• PR description contains enough context, bug link, code samples for new APIs
• Tested (unit tested)
• Docs/readmes updated (README already documents visualDetailLevel; no doc change)

Decision

Strip visualDetailLevel in the existing destructure list. Alternative: also add it to ChatCompletionCommonConfigSchema, or share a denylist with the image/audio/translate converters.

Stripping in the destructure matches toOpenAIRequestBody()'s current pattern and the Go adapter (go/plugins/compat_oai/generate.go already deletes visualDetailLevel). Putting it on the Zod schema would not stop the leak because common config is .passthrough(). Can switch to the schema or shared-denylist approach if that is preferred.

Test plan

• Regression: toOpenAIRequestBody('gpt-4o', { config: { visualDetailLevel: 'high' }, ...image... }) has no top-level visualDetailLevel and still sets image_url.detail to 'high'.
• Test fails without the source change (Received value: "high" on not.toHaveProperty('visualDetailLevel')) and passes with it.
• @genkit-ai/compat-oai unit tests (7 suites, 102 tests).

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[gemini-code-assist]

Code Review

This pull request extracts visualDetailLevel from the request configuration to prevent it from leaking into the OpenAI request body, and adds a test to verify this behavior. The feedback recommends adding visualDetailLevel to ChatCompletionCommonConfigSchema to provide schema validation, type safety, and autocompletion, while also utilizing the otherwise unused VisualDetailLevelSchema.

[gemini-code-assist]

While destructuring visualDetailLevel here correctly prevents it from leaking into the OpenAI request body, visualDetailLevel is currently missing from ChatCompletionCommonConfigSchema (defined on line 79). As a result, users do not get autocompletion, type safety, or schema validation for this option. Additionally, VisualDetailLevelSchema (defined on line 70) remains unused dead code.\n\nTo resolve this, consider adding visualDetailLevel to ChatCompletionCommonConfigSchema in js/plugins/compat-oai/src/model.ts:\n\ntypescript\nexport const ChatCompletionCommonConfigSchema =\n GenerationCommonConfigSchema.extend({\n temperature: z.number().min(0).max(2).optional(),\n frequencyPenalty: z.number().min(-2).max(2).optional(),\n logProbs: z.boolean().optional(),\n presencePenalty: z.number().min(-2).max(2).optional(),\n topLogProbs: z.number().int().min(0).max(20).optional(),\n visualDetailLevel: VisualDetailLevelSchema,\n });\n