CC: Improve performance of responses to user

[matthewbolanos]

The goal of this task is to speed up the initial time for feedback after the user has sent a request. Below is a list of optimizations that could be made to the application. We should continue making improvements to the experience until time-to-first message is under 5 seconds.

• Parallelizing LLM calls
• Stream responses from the bot
• Provide a "bot is typing" message
• Provide messaging describing what the bot is doing (e.g., creating plan, performing step 1, step 2, etc.)
• Print an initial message back to the user before the plan is fully crated.

[matthewbolanos]

Is there a reason why we're using prompts for ExtractAudienceAsync? Doing that with native code could help speed things up and avoid using tokens.