fix(sessions): drop empty reasoning items from OpenAIConversationsSession persistence

[xodn348]

Summary

When using OpenAIConversationsSession with a reasoning-enabled model (e.g. gpt-5.4-mini with reasoning: low), the SDK can produce reasoning run items whose summary list is empty. The session persistence path converted these to {"type":"reasoning","summary":[]} and attempted to save them via conversations.items.create(...), which the Conversations API rejects with Invalid item.

This PR fixes that by:

• Adding _is_unpersistable_for_openai_conversation() — a targeted predicate that returns True for {"type":"reasoning","summary":[]} items.
• In save_result_to_session, filtering such items from the add_items payload after the persisted-item counter is updated. The counter still advances past empty reasoning items so the retry/streaming deduplication logic works correctly; we just stop sending them to the API.

Non-empty reasoning items (with at least one summary entry) are unaffected and continue to be persisted normally.

=== LOCAL_TEST_PASSED ===

Test plan

Two new pytest tests added to tests/test_agent_runner.py:

• test_save_result_to_session_drops_empty_reasoning_from_openai_conversations — verifies that an empty reasoning item is not sent to add_items but is counted in saved_run_items_count and _current_turn_persisted_item_count.
• test_save_result_to_session_keeps_nonempty_reasoning_in_openai_conversations — regression guard confirming that reasoning items with a non-empty summary still reach add_items.

uv run ruff check src/agents/run_internal/session_persistence.py tests/test_agent_runner.py
uv run pytest tests/test_agent_runner.py -k "session" -v
uv run pytest tests/memory/ -v

All 45 session tests and 119 memory tests pass.

Issue number

Closes #3268

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've run make lint and make format
• I've made sure tests pass