langchain: streamed calls are not marked as streaming, thus no streaming timing signals emitted

[sfc-gh-zeningchen]

Describe your environment

OS: Ubuntu 22.04
Python version: Python 3.12
Package version: opentelemetry-instrumentation-genai-langchain 1.1b1, opentelemetry-util-genai 1.1b0 (also reproduces on main)
GenAI library (e.g. anthropic, openai) and version: langchain 1.3.16, langchain-openai 1.4.3

What happened?

opentelemetry-util-genai 1.1b0 added streaming telemetry in #269 — the
gen_ai.request.stream and gen_ai.response.time_to_first_chunk span
attributes, and the gen_ai.client.operation.time_to_first_chunk and
gen_ai.client.operation.time_per_output_chunk histograms. None of it reaches
the LangChain path.
Those values are set by the shared stream wrappers, which need an SDK stream to
wrap. The LangChain instrumentation is callback-based: it opens an
InferenceInvocation in on_chat_model_start, finishes it in on_llm_end, and
never sees a stream. on_llm_new_token is not implemented, so there is no
per-chunk hook either.
A streamed call is therefore indistinguishable from a synchronous one, and
time-to-first-token cannot be separated from total duration.

Steps to Reproduce

Stream a chat call with the instrumentation enabled:

LangChainInstrumentor().instrument()
list(ChatOpenAI(model="gpt-4o-mini", stream_usage=True).stream("Say hello."))

Then inspect the finished span's attributes and the emitted metric names.

Expected Result

The chat span carries gen_ai.request.stream = true and a plausible
gen_ai.response.time_to_first_chunk, and both streaming histograms record
points. Non-streamed calls are unchanged.

Actual Result

Both span attributes are absent and neither histogram is emitted.

Additional context

No response

Would you like to implement a fix?

Yes

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}