Update 05- sample, sticking with generators here because they give us a lot more control over the order that things get queued.

src/samples/theoretical-to-real/04-utterance-and-speech.py

@@ -32,6 +32,10 @@ async def main(room_url:str):
         # queue two pieces of speech: one specified as a text literal,
         # and one generated by an llm. We'll kick off the llm first, and let
         # it generate a response while we're speaking the literal string.
+        #
+        # Note that in this case, we don't use `run_llm_async` because we're
+        # taking advantage of the time spent speaking the first phrase to generate
+        # the entire LLM response, and this happens asynchronously in a task.
         llm_response_task = asyncio.create_task(llm.run_llm(
             [{"role": "system", "content": "tell the user a joke about llamas"}]
         ))