feat(llma): send number of web searches

[Radu-Raicea]

This now track web search tool calls across various providers, and sends the web search count as an event property ($ai_web_search_count) to be used in calculating the LLM costs, where applicable.

[greptile-apps]

_{8 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[chatgpt-codex-connector]

💡 Codex Review

posthog-python/posthog/ai/openai/openai_converter.py

Lines 417 to 419 in 312b9c8

	if provider_type == "chat":
	if not hasattr(chunk, "usage") or not chunk.usage:
	return usage

Detect web searches before early exit in streaming chunks

For chat completions, extract_openai_usage_from_chunk immediately returns when a chunk lacks usage. The call to extract_openai_web_search_count() is placed after this guard, so any streaming chunk that contains web search indicators (citations, web_search_call, etc.) but no usage is ignored and $ai_web_search_count will never be reported. OpenAI can emit citations before the final chunk where token usage is provided, meaning web searches won’t be captured for such streams. Extract the web search count before the early return or separate it from the usage check so non‑usage chunks can still contribute to the count.

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".

[andrewm4894]

lgtm apart form merge conflict but approving so no delay once that sorted

[Radu-Raicea]

@andrewm4894 Thank you!

For everyone: please do not merge, I need to run a couple of sanity checks before I merge.