Action counts modify policy state

[jdherman]

The idea was for a policy to save its own frequency of actions during a simulation (e.g. "Flood Control (5%)"). However this seems to create a number of problems:

• During parallelization, the policies themselves need to be returned to master because they are modified when running the objective function

• After snapshots are saved, if a policy is run again in simulation mode, it does not clear the action count. This results in frequencies that sum to more than 100%. Should be an easy fix.

• The original goal was to prune actions that never occur. But I never got around to doing that anyway.

I'm wondering if it would be easier to just remove this altogether. Action counts are not a property of the policy anyway, they depend on the forcing scenario. Might be better to leave this up to the simulation model to compute, outside of ptreeopt.