Possible zero downtime supervisor reloads

[ankush]

We use --preload so it's not really possible to reload code in running servers.

BUT gunicorn supports --reuse-port which means there can be two gunicorn instances listening on same kernel socket. So theoretically we can do:

1. Update code (or not, it can just be a routine restart for scaling workers)
2. Start a new gunicorn version on same port.
3. Once step 2 completes, send terminate signal to original one.

This is possibly complex to do with supervisord limitations.

[ankush]

Eh, over engineering. But maybe someday when we aren't using supervisord