Improve how we start and run the app in production

[reefdog]

We currently fire the app up in production with pm2 start 'yarn start', which may end up creating more processes and daemons than we need in production. E.g., we definitely don't need parcel running (#334), and should probably not be using nodemon either. This should probably just be pm2 start src/server/index.js or something.

Anyway just spit-balling; the point is we're probably not starting our app the right way and this may be the source of some of our pm2 management problems!

[slifty]

I do think we should probably have it be yarn start or something similarly black-box-expected-to-start-things-properly. However I think yarn start needs fixing as described in #334.

[slifty]

One note: we have a scripts/startup.sh for CD -- it is only being invoked in dev.

We can modify the pm2 script there quite easily. I'd like to tease out our options based on this issue:

1. Have the script use yarn start (this is status quo)
2. Modify the script to directly call node
3. Create a new yarn command yarn start:prod

My concern with 2 is that it creates two places we would have to update if we added startup dependencies, and it's unclear what it accomplishes.

My concern with 3 is that I think it would be identical to the current yarn start.

Thoughts?