feat: trace hosted web search in Langfuse

[AlexZahar]

Summary

• request web_search_call.action.sources for OpenAI and Azure Responses calls using hosted web search
• emit each provider web_search_call as one child Langfuse tool observation
• record the action, provider call ID, status, source count, and source URLs without fetched page bodies
• deduplicate repeated streamed output items by provider call ID
• keep telemetry best-effort so tracing failures cannot block model responses

This is independent from #317. It does not change web-search routing or execute an additional search.

Verification

• npm test -- --runInBand src/llm/openai/llm.spec.ts src/llm/openai/contentBlocks.test.ts src/specs/langfuse-callbacks.test.ts (71 passed)
• npm run build
• Prettier, ESLint, and staged-file checks passed.

[danny-avila]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 80d3a64f95

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Include versioned hosted web-search tools

With the current Responses API, hosted web search tools can also be sent as web_search_2025_08_26, not just web_search/web_search_preview (OpenAI API reference). In that case this predicate returns false, so invocationParams never adds web_search_call.action.sources and the Langfuse tool observation records no sources even though the call used hosted web search. Include the versioned hosted web-search tool type here.

Useful? React with 👍 / 👎.