Add cpu utilization as a parameter with which desires can be obtained

[ramsrivatsa]

Use CPU utilization directly within extra model arguments. If we have utilization numbers, we can directly leverage them to obtain desired instance types. If we don't, then use the default reads and writes to compute CPU util which in turn can be used to obtain desired instance types.

Also added pytest in the setup so that it can integrate seamlessly with pycharm.

[abersnaze]

i noticed a lot of whitespace change. @jolynch what is there a formatter that you used?

[abersnaze]

I don't think the model should be aware what the current instance is.

I would like to see the current instance to be used as signal to the planner that one of the least_regret plans must be of that instance. that way we could:

• compare what the planner thinks should deploy vs what is.
• compare replans vs alternatives.

this also introduces more challenges to compositions of models since a kv cluster would likely have a different current instance than the composed cassandra. I'm thinking we should depreciate composition and have the client invoke the planner for each model separately.