[Feature Request] Make LocalLauncher support launching perpetual serving

[logan-markewich]

Right now, LocalLauncher only supports launch_single(), which is kind of a one-shot setup, run, teardown process.

In addition, I think there's room for a pseudo launch_server() that runs everything in a single async process.

I've seen at least two people now trying to run the server launcher inside of an already existing fast-api server, which won't work. Currently you'd need to launch llama-agents separately from your existing fast-api server and use a client to communicate.