Experiment server

[sondreso]

Issue
Resolves #1250

Approach

1. Refactor out the tracking of an evaluation.
2. Allow the evaluator to track multiple evaluations
3. Add the concept of an experiment with its own status, containing multiple evaluations.
4. Make the run models setup, start and finish the experiment at the correct times.
5. Make experiment endpoint for clients (decide on "out-of-band" communication for experiments and evaluations vs all status in one channel)
6. Make evaluator communication session
7. Cleanup

[sondreso]

Will reopen a new pull request in the future