fix again vllm for non base64 (#818)

# What does this PR do?

- previous fix introduced regression for non base64 image
- add back download, and base64 check


## Test Plan

<img width="835" alt="image"
src="https://github.com/user-attachments/assets/b70bf725-035a-4b42-b492-53daaf71458a"
/>


## Sources

Please link relevant resources if necessary.


## Before submitting

- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Ran pre-commit to handle lint / formatting issues.
- [ ] Read the [contributor
guideline](https://github.com/meta-llama/llama-stack/blob/main/CONTRIBUTING.md),
      Pull Request section?
- [ ] Updated relevant documentation.
- [ ] Wrote necessary unit or integration tests.

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -177,7 +177,8 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
         if isinstance(request, ChatCompletionRequest):
             if media_present:
                 input_dict["messages"] = [
-                    await convert_message_to_openai_dict(m) for m in request.messages
+                    await convert_message_to_openai_dict(m, download=True)
+                    for m in request.messages
                 ]
             else:
                 input_dict["prompt"] = await chat_completion_request_to_prompt(