Manage state information more cleanly

Currently, the method state information is mostly stored in the candidate space object. All state information could be moved into a separate state object, such that the candidate space is no longer stateful. e.g. there are a few objects (Problem, ModelSubspace, ModelSpace, CandidateSpace) that each contain some logic to manage exclusions -- this could all be moved to the state object.

State information (some of which is only used by the FAMoS method currently):

• each iteration:
  • the identified models
  • the calibrated models
  • the method
• exclusions (accumulative over iterations)

Partially resolved by #128 , where all calibrated models from all iterations are now stored in Problem.state.

Use this in plotting code too, e.g.:

• add method information from each iteration to graph_iteration_layers, to make FAMoS method switches clear

Start by tracing state and exclusion logic across Problem, ModelSubspace, ModelSpace, and CandidateSpace, including the state changes from #128. Then inspect graph_iteration_layers and Problem.state to determine how iteration methods should be represented. Done means state is managed separately from CandidateSpace and plotting makes FAMoS method switches clear.