Fix sliding window cutoff logic (#9261)

* fix sliding window cutoff calculations to use agent instead of summarizer config

* allow approval messages with tool_calls as valid cutoffs, prevent approval pairs from being split

* update tests with updated sliding window parameters

---------

Co-authored-by: Amy Guan <amy@letta.com>

tests/integration_test_summarizer.py

@@ -1157,6 +1157,7 @@ async def test_sliding_window_cutoff_index_does_not_exceed_message_count(server:
         summary, remaining_messages = await summarize_via_sliding_window(
             actor=actor,
             llm_config=llm_config,
+            agent_llm_config=llm_config, # case where agent and summarizer have same config
             summarizer_config=summarizer_config,
             in_context_messages=messages,
         )