API key endpoint (/openai/v1) does not support Responses API for non-OpenAI models

## Summary

The account-level API key endpoint (`/openai/v1`) does not support the Responses API for non-OpenAI model formats (e.g., DeepSeek-R1-0528). Calls return `This model is not supported by Responses API.` Only OpenAI-format models (e.g., gpt-4.1-mini) work with the Responses API on this endpoint.

The project-level EntraID endpoint (`/openai` with `api-version`) supports the Responses API for **all** model formats including non-OpenAI models.

## Steps to Reproduce

1. Deploy DeepSeek-R1-0528 on a Foundry account
2. Get the API key via `az cognitiveservices account keys list`
3. Call the Responses API on `{account-endpoint}/openai/v1` with model `DeepSeek-R1-0528`

## Expected Behavior

The Responses API call should succeed, as DeepSeek-R1-0528 supports the Responses API on the project endpoint.

## Actual Behavior

`This model is not supported by Responses API.`

## Impact

This means API key authentication cannot be used as a universal approach when the template includes non-OpenAI models. The template uses **EntraID authentication only** to avoid this limitation.

## Workaround

Use EntraID (`DefaultAzureCredential`) authentication with the project endpoint (`AZURE_AI_PROJECT_ENDPOINT` + `/openai`) instead of API key authentication.