To make Everserver more explicit, we should split running Everest into two parts:

1. Submitting and starting server
2. Starting experiment

Both of these should be done without requiring an Everest config on disk, i.e. we should start the experiment through a new endpoint on the server, which takes the serialized experiment config.

We can still create ErtConfig server side for now, can change that in a later step.

Everserver should still shut down once the experiment has completed like it does currently. We can change that behavior in a later step.