fix(sessions): do not forward session limit as Conversations page size

[tiagovilasboas]

Summary

This pull request fixes OpenAIConversationsSession.get_items(limit=N) forwarding the SDK history cutoff as conversations.items.list(limit=N).

For a large finite session limit that exceeds the Conversations API page-size constraint, an otherwise valid Session.get_items(limit=N) call could fail before the OpenAI Python client paginated. Maintainer guidance on #4721 was to omit limit from the provider list call, let the client own pagination, and keep the SDK cutoff local.

The finite-history path now requests items in descending order without a provider page-size argument, stops locally once N items are collected, and reverses them into chronological order. Unlimited and zero-limit paths are unchanged.

Test plan

• Added a regression that requests 101 newest-first items and asserts:
  • conversations.items.list() is called without limit
  • collection stops after exactly 101 items
  • the returned list is the newest 101 items in chronological order
• uv run pytest -q tests/memory/test_openai_conversations_session.py tests/memory/test_session_limit.py — 56 passed
• Independent ordinary review of the complete merge-base diff — clean; no findings
• .agents/skills/code-change-verification/scripts/run.sh:
  • make format — passed
  • make lint — passed
  • make typecheck — passed (mypy + pyright, 0 errors)
  • make tests — 9611 passed, 29 skipped (parallel); 77 passed, 4 skipped, 55 deselected (serial)

Issue number

Fixes #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