fix(server): preserve root agent activity

[tarik02]

What Changed

• Filter child conversation notifications only for provider thread IDs learned from collabAgentToolCall or subAgentActivity.
• Document that root-agent activity stays visible.

Why

The provider-thread mismatch heuristic could classify a resumed root thread as a child. That kept allowlisted commands and file changes while dropping root messages and reasoning. Explicit child relationships already identify the notifications that need filtering.

Checklist

• This PR is small and focused
• I explained what changed and why

gpt-5.6-sol via T3 Code.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5a107c7531

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]

Add regression tests for root and child notification routing

This changes backend notification classification without adding focused coverage; the existing CodexSessionRuntime.test.ts suite does not exercise root-versus-child routing. Add cases proving that a mismatched resumed-root thread retains messages/reasoning while explicitly discovered child threads suppress conversation events but retain allowed activity, so this fix cannot silently reintroduce either regression.

AGENTS.md reference: AGENTS.md:L101-L101

Useful? React with 👍 / 👎.