Don't choke on Azure OpenAI content filtering metadata

[laszlovandenhoek]

On Azure OpenAI, when streaming responses, the first chat completion chunk may contain an empty choices[] array, and instead only contains information about filtered categories:

{"id":"","choices":[],"created":0,"model":"","object":"","system_fingerprint":null,"prompt_filter_results":[{"prompt_index":0,"content_filter_results":{"hate":{"filtered":false,"severity":"safe"},"self_harm":{"filtered":false,"severity":"safe"},"sexual":{"filtered":false,"severity":"safe"},"violence":{"filtered":false,"severity":"safe"}}}]}

This is the same issue as microsoft/semantic-kernel#3650.

Directly indexing into chunk.choices[0], as the Mentat code currently does, naturally causes an IndexError: list index out of range.

I'll submit a PR to rectify this shortly.