Skip api.openai.com/v1/models refresh when using custom responses-api-endpoint

Problem

When using codex-action@v1.4 with a custom responses-api-endpoint (e.g., an in-house OpenAI-compatible proxy), the Codex CLI still calls https://api.openai.com/v1/models on startup to refresh available models.

If the API key is issued for the custom endpoint and not for OpenAI directly, this call always fails with a 401 Unauthorized:

ERROR codex_core::models_manager::manager: failed to refresh available models:
unexpected status 401 Unauthorized: Missing bearer authentication in header,
url: https://api.openai.com/v1/models?client_version=0.98.0

The error is non-fatal (the CLI continues), but it's noisy and confusing in CI logs.

Expected behavior

When a custom responses-api-endpoint is configured, the CLI should either:

• Skip the api.openai.com/v1/models refresh entirely, or
• Call the custom endpoint's /v1/models instead, or
• Provide a flag to disable the model refresh (e.g., --skip-model-refresh)

Reproduction

1. Configure codex-action@v1 with a custom responses-api-endpoint and an API key that is valid only for that endpoint
2. Run the action
3. Observe the 401 error in logs from the direct api.openai.com/v1/models call

Environment

• Codex CLI v0.98.0
• openai/codex-action@v1
• Using responses-api-endpoint pointed at an in-house proxy