Unintuitive switching behavior of Meta-Recommenders in combination with Campaign

[Scienfitz]

According to the current logic, meta-recommenders perform the logic that decides the switching at the point of the .recommend call. However, intuitively imo it would make sense to perform that logic at the time of .add_measurements, simply because adding the measurements marks the end of an iteration, and meta-recommenders switch based on iterations.

The current behavior requires the following unintuitive workaround:

recommender = TwoPhaseMetaRecommender(
    initial_recommender=RandomRecommender(),
    recommender=BotorchRecommender,
    switch_after=1,
) 

campaign = Campaign(..., recommender=recommender)
recs = recommender.recommend(5)
### perform measurements
campaing.add_measurements(my_measurements)

# Analyze the surrogate based on the first set of measurements
surrogate = campaign.get_surrogate()

This will fail with RuntimeError: The current recommender is of type 'RandomRecommender'. Surrogate models are only available for recommender subclasses of 'BayesianRecommender'. because the meta recommender has not yet advanced the recommender to the second one. This requires another emtpy call to .recommend, which would not be needed anymore in the suggested change.