fix(sessions): keep Conversations history limits local

[sylvesterkaczmarek]

Summary

This pull request fixes large OpenAIConversationsSession.get_items(limit=N) calls by keeping the SDK history limit local instead of forwarding it as the Conversations API page-size parameter.

The previous #4721 established the underlying failure, but its hard-coded provider page-size cap was closed in favor of a narrower maintainer-requested approach: omit limit from conversations.items.list(), let the OpenAI client handle provider pagination, stop locally once the requested number of items has been collected, then reverse the descending results back into chronological order.

This preserves the existing Session.get_items(limit=N) behavior without duplicating provider-owned pagination limits in the Agents SDK.

Test plan

• Added a focused regression that requests 150 items from a 200-item descending stream.
• Verifies the provider call receives no limit argument.
• Verifies the SDK still returns exactly the newest 150 items in chronological order.

Issue number

Closes #4757

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-08-29T15:50:36.113937Z	2b147b7	PR opened
🔒 Security Review	✅ Completed 2026-08-29T15:55:27.958848Z	2b147b7	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[linhongyu510]

I independently checked the pagination contract on 2b147b7f. The installed OpenAI SDK AsyncCursorPage.__aiter__() iterates iter_pages(), which calls get_next_page() until exhaustion, so omitting the provider limit still traverses multiple provider pages while this loop stops locally at the resolved Session limit. The descending collection plus final reverse also preserves the Session contract of returning the newest N items in chronological order.

I ran the new pagination regression together with the existing tests/memory/test_openai_conversations_session.py suite from the PR snapshot: 39 passed. I found no blocking behavior issue in this change. This is independent local evidence only; the PR currently has not marked the repository verification checklist as completed.

[seratch]

This addresses the same paging bug as #4873 with the same runtime change. I recommend consolidating on #4873, whose regression also checks that collection stops at the requested count, and closing this duplicate. The intended contract remains newest-N Session items returned chronologically across API pages.