Two new methods: from_toy_hdf and from_plain_dict (counterpart to to_plain_dict, used inside from_toy_hdf).
Problems:
to_plain_dict (fit-initial not saved to plain_dict) and back (hard to recover flattened structure, highly order dependent code) is not bijektive at the moment and probably hard to make (flat structure is required for DataFrame/HDF).
additionally, the toy fit returns more values than other results do (comparing to Roofit) like fitnum/jobid/fit_strategy/model_name. Should we may just store them as well and alter converters (to_yaml etc, but probably not needed) and getters to obtain this values if available?