SMAC support should incorporate runhistory

[bpkroth]

trafficstars

SMAC has an internal notion of runhistory that we currently seem to ignore when re-initializing the optimizer. This affects several things:

1. During initialization we always perform a large number of random initializations because we haven't told it we're about to prewarm it with historical data.
2. During the "intensification" it doesn't have the history to know what incumbents should be retried and compared against.