feat:Add token metrics support to autogen-agentchat instrumentation with streaming support

[fenilfaldu]

Closes #2258
Before

After

---

Note

Capture LLM model name and detailed token usage in BaseOpenAIChatCompletionClient.create and .create_stream, with streaming usage injection and end-of-stream attribute setting.

• Instrumentation: OpenAI ChatCompletion (_wrappers.py)
  • LLM token metrics: Extract token usage from CreateResult.usage (prompt, completion, total) with details (prompt: cache_read/audio/cache_input; completion: reasoning/audio) and set via get_llm_token_count_attributes, plus explicit reasoning/audio completion detail attributes.
  • Model metadata: Add model name via get_llm_model_name_attributes to LLM spans.
  • Streaming support: Ensure include_usage (via extra_create_args.stream_options or include_usage param); during streaming, accumulate output/tool-call/token attributes and set them after stream completion.
  • Outputs/Tools: Continue setting output attributes and extracting tool call attributes from responses.

^{Written by Cursor Bugbot for commit ab0a81a. This will update automatically on new commits. Configure here.}

[cursor]

Bug: Token Attribution Fails for Cached Input

The mapping includes "cache_input": "text_tokens", but get_llm_token_count_attributes doesn't handle the cache_input key in prompt_details. If text_tokens exists in the response, it gets extracted into the token usage dict but is never converted to a span attribute, making this extraction ineffective.