fix(azure-ai): fall back to message metadata for gen_ai.response.id

[DevNinja (0xDevNinja)]

Fixes #655.

Problem

AzureAIOpenTelemetryTracer.on_llm_end only stamps gen_ai.response.id when it finds LLMResult.llm_output["id"]. That field is populated on the Chat Completions path, but on the Responses API path langchain-openai exposes the id only on AIMessage.response_metadata["id"]. As a result gen_ai.response.id is missing on that path, and the traces do not show up in the Foundry portal.

Fix

Resolve the response id with a fallback: use llm_output["id"] when present, otherwise walk the chat generations and read AIMessage.response_metadata["id"]. The attribute is now emitted on both paths, and llm_output["id"] still takes precedence.

response_id = llm_output.get("id")
if not response_id:
    for gen in chat_generations:
        message = getattr(gen, "message", None)
        metadata = getattr(message, "response_metadata", None)
        if isinstance(metadata, Mapping):
            response_id = metadata.get("id")
            if response_id:
                break
if response_id:
    record.span.set_attribute(Attrs.RESPONSE_ID, str(response_id))

Tests

Added two unit tests in tests/unit_tests/test_inference_tracing.py:

• test_response_id_falls_back_to_message_response_metadata — id only on AIMessage.response_metadata is now emitted.
• test_response_id_prefers_llm_output_over_message_metadata — llm_output["id"] wins when both are present.

Full test_inference_tracing.py module passes (138 tests); ruff + mypy clean.

Fixes #655