Added span support for genAI langchain llm invocation

[wrisa]

Description

Taking inspiration from traceloop's openllmetry support for langchain instrumentation, added callback handler support for llm start and end to generate span with attributes(no content) listed here.

Attributes added on span:
-> gen_ai.operation.name: Str(chat)
-> gen_ai.request.model: Str(gpt-3.5-turbo)
-> gen_ai.request.top_p: Double(0.9)
-> gen_ai.request.frequency_penalty: Double(0.5)
-> gen_ai.request.presence_penalty: Double(0.5)
-> gen_ai.request.stop_sequences: Slice(["\n","Human:","AI:"])
-> gen_ai.request.seed: Int(100)
-> gen_ai.request.max_tokens: Int(100)
-> gen_ai.provider.name: Str(openai)
-> gen_ai.request.temperature: Double(0.1)
-> gen_ai.response.finish_reasons: Slice(["stop"])
-> gen_ai.response.model: Str(gpt-3.5-turbo-0125)
-> gen_ai.response.id: Str(chatcmpl-Bz8yrvPnydD9pObv625n2CGBPHS13)
-> gen_ai.usage.input_tokens: Int(24)
-> gen_ai.usage.output_tokens: Int(7)

Please refer complete span attributes o/p: https://docs.google.com/document/d/13SXuf4iZSl2MqWUJtmoDEqFxuHZ1Y8hykW_ctlykcDE/edit?tab=t.0

We have just added span support in this PR for now to keep this PR concise and in future we will have metric and log support PRs.

Fixes # (issue)

Type of change

Please delete options that are not relevant.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration

• From test folder /opentelemetry-python-contrib/instrumentation-genai/opentelemetry-instrumentation-langchain/tests run pytest -rP test_llm_call.py

Does This PR Require a Core Repo Change?

• Yes. - Link to PR:
• No.

Checklist:

See contributing.md for styleguide, changelog guidelines, and more.

• Followed the style guidelines of this project
• Changelogs have been updated
• Unit tests have been added
• Documentation has been updated

[pmcollins]

Thanks for putting this together. Added some comments.

[aabmass]

My biggest concern is if this is leaking memory otherwise just some small requests

[wrisa]

Action items from Aug 28 2025 Python SIG:

• Explain value of LLM Invocation in LangChain vs client-instrumentation (i.e. openai)
  Added aws bedrock support as its specific instrumentation is not yet available
• Unit-test showing both openai/bedrock providers to match telemetry
  Added aws bedrock support
• How TraceLoop/OpenInference do it, document their approach
  It has both langchain and openai instrumentation available but it produces only langchain span. Uses SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY in context_api.
• Opt-in telemetry for a short term and long-term strategy to avoid telemetry duplication.
  Add similar key SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY = "suppress_language_model_instrumentation" in semantic conventions. Implement logic to set value true in langchain instrumentation and later add support to check in downstream instrumentation(eg:openai-v2) if value already true then skip telemetry
• we’ll provide both testing with another llm and opt-in for filing attributes.
  WIP

[wrisa]

Sept 9 2025 SEMCOV meeting discussion.

• Opt-in telemetry for a short term and long-term strategy to avoid telemetry duplication.
  [Aaron] - We can put SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY into genai-util
  OpenAI instr could react and not instrument.
  [Liudmila] - TracerConfig for disabling https://opentelemetry.io/docs/specs/otel/trace/sdk/#tracerconfig,
  https://opentelemetry.io/docs/specs/otel/trace/sdk/#tracerconfigurator takes the full InstrumentationScope
  example config - https://github.com/open-telemetry/opentelemetry-configuration/blob/094671ba39b016a992d11a32958c7759b244aeb3/examples/kitchen-sink.yaml#L776
  AI: Liudmila will create issue in semconv, discuss if scope name ot attributes are more appropriate for filtering

• we’ll provide both testing with another llm and opt-in for filing attributes.
  Currently ChatOpenAI and ChatBedrock are supported and telemetry for other providers/LLM are skipped and they will be supported in future. Test added for unsupported(gemini)