Re-architect code to use Redis for control messages #23

Atleast 2 re-architecturing changes come to mind:

• remote_monitor should not be orchestrating PrometheusClient. It is only doing right now so that it knows when to start and stop recording resource usage.
• PrometheusClient should not be orchestrating profilers. It is only doing so right now so that it knows when to start and stop profiling

Both these usecases (and others) should be solved by the "master orchestrator" (currently, experiment_run_e2e.py) sending control messages to various components