Errors from custom model providers dont handled

[TypeHintsFun]

When custom model provider sends an error message (e.g. rate limit exceeded), agents falls with error

...\agents\models\openai_chatcompletions.py", line 134, in get_response
    f"LLM resp:\n{json.dumps(response.choices[0].message.model_dump(), indent=2)}\n"
                             ~~~~~~~~~~~~~~~~^^^
TypeError: 'NoneType' object is not subscriptable

Im use an gemini model through openrouter, and answer looks like this:

{
  "id": null,
  "choices": null,
  "created": null,
  "model": null,
  "object": null,
  "service_tier": null,
  "system_fingerprint": null,
  "usage": null,
  "error": {
    "message": "Rate limit exceeded: google/gemini-2.0-flash-exp/...",
    "code": 429,
    "metadata": {
      "headers": {
        "X-RateLimit-Limit": "4",
        "X-RateLimit-Remaining": "0",
        "X-RateLimit-Reset": "1743142560000"
      },
      "provider_name": "Google AI Studio"
    }
  },
  "user_id": "..."
}

As i can see, agents dont have any provider errors handling mechanism (including errors from OpenAI), so it would be nice to add it

[seratch]

We will track the progress at #604