feat: Last9LogToSpanProcessor — bridge GenAI log events onto spans (v1.1.0)

[prathamesh-sonpatki]

Summary

• Adds Last9LogToSpanProcessor, a new LogRecordProcessor that promotes
  GenAI log events emitted by opentelemetry-instrumentation-openai-v2 onto
  the active span as flat + indexed attributes + span events so the Last9
  LLM dashboard renders prompts, completions, and tool calls.
• Last9SpanProcessor gains an optional log_processor= kwarg for per-span
  counter cleanup.
• Bumps version to 1.1.0.

Why

opentelemetry-instrumentation-openai-v2 (v2+) follows the new OTel GenAI
semantic conventions — it emits message content, tool calls, and completions
as OTel log events, not as span attributes. The Last9 LLM dashboard reads
span attributes (gen_ai.prompt, gen_ai.completion) and span events
(gen_ai.content.prompt, gen_ai.content.completion), so those payloads
were invisible until now.

This bridge closes the gap without monkey-patching the openai SDK. Works for
any framework that routes LLM calls through openai-v2 (AutoGen, LangChain,
direct usage).

Test plan

• Unit test suite: uv run pytest tests/ -x -q — 117 passing.
• End-to-end demo with AutoGen AssistantAgent + tool call; chat span
  carries gen_ai.prompt, gen_ai.completion, gen_ai.content.prompt
  event, and gen_ai.content.completion event.
• Unit tests for Last9LogToSpanProcessor itself (follow-up — coverage
  currently 21%, functionally verified via integration demo).

Notes

Python 3.14 users must pin wrapt<2 because
opentelemetry-instrumentation-openai-v2 2.3b0 calls
wrap_function_wrapper(module=..., name=..., wrapper=...) and wrapt 2.0
renamed the first kwarg to target=. Without the pin, instrumentation fails
silently. README + CHANGELOG call this out.

Not in scope

• Tool-call argument / result capture on AutoGen's own execute_tool span
  (tracked as Phase 2 — requires wrapping autogen_core.tools.FunctionTool.run_json).
• Anthropic / google-genai bridges (deferred per customer priority).

🤖 Generated with Claude Code