Use ERT storage for Everest simulator cache

[yngve-sk]

The current Everest simulator cache keeps a mapping of control values -> results, where results are ERT responses for objective/constraint values.

Storing these objective/constraint values separately is redundant, as we already store this in ERT storage. This mapping could be refactored to use ERT storage, so that we can instead keep a mapping of control values -> [(batch, realization), ...], where we say which realizations of which batch contain simulation results, given a set of control values. This setup would be less redundant, and let us easier traceback if some results are cached, and where they originate from.