Potential fix for non responsive LLM after summary

[Leidtier]

This wraps openai_client.streaming_call() and openai_client.request_call() to be behind the same lock, blocking any of them being called while a call to either of them is still ongoing.

The fix is potential because:

• I wasn't able to find the exact reason why calls to openai_client.streaming_call() didn't resturn any result.
• I simply assumed that there is a possibility that both can be called in parallel.
• Since this was implemented I didn't get another block in my tests, but because of the randomness of this, I can't be sure.
• Having more locks has the potential of unwanted deadlocks, but I didn't encounter any.