[Data][LLM] Support OpenAI's nested image_url schema in PrepareImageStage

[GuyStone]

Why are these changes needed?

• Add support for OpenAI-compatible nested image_url schema in PrepareImageStage.
• Currently, the documentation and code indicate that inputs should follow the OpenAI chat messages format. However, the accepted image format differs from what OpenAI supports. This inconsistency creates confusion and increases complexity when switching between Batch OpenAI and Batch with Ray.

The first stage of the processor is ChatTemplateStage.
Required input columns:
        messages: A list of messages in OpenAI chat format. See https://platform.openai.com/docs/api-reference/chat/create for details.

Related issue number

• N/A

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[gemini-code-assist]

Code Review

This pull request adds support for OpenAI's nested image_url schema in PrepareImageStage, which is a great improvement for consistency and user experience. The implementation is clean and correctly handles the new format, including validation for malformed inputs. A new unit test is also added to cover the happy path. My only suggestion is to also add a test case for the error path to ensure the new validation logic is fully covered.

[gemini-code-assist]

This is good validation to ensure the image_url dictionary is well-formed. To make this more robust, it would be beneficial to add a unit test case that verifies this ValueError is raised when the url key is missing. This would improve test coverage for error handling paths.

[nrghosh]

Supporting nested image_url makes perfect sense, thanks for the contribution @GuyStone

Few quick nits

• there's an optional param called detail (From openai schema that we're dropping here. I don't think the downstream templates / etc would currently use it, but would be good to just leave a note in the docstring that we're not passing it forward
• Would be good to have at least one or two unhappy path tests - ex. missing or non-string image_url
• type-guard + clarify error message, something like

if content_item["type"] == "image_url" and isinstance(image_data, dict):
    url = image_data.get("url")
    if not isinstance(url, str) or not url:
        raise ValueError("image_url must be an object with a non-empty 'url' string")
    image = url

otherwise looks good - kicked off a release test in the meantime

[nrghosh]

LGTM after review

Not re-running release tests, so good to merge cc @kouroshHakha