Do we need a generic "unreportable" state?

[tms-bananaquit]

• In the Maciel 2015-style ensemble, some detectors have their output ignored while they "wait" and learn a potential new concept.
• In MD3, the waiting_for_oracle status also looks like one where it may be appropriate for something downstream of the detector to ignore its output.
  • This is sort of a weirdo in that we don't have a conclusive treatment on how new samples should be passed to semi-supervised detectors. Should they be coming from the oracle, or should they be coming from a normal stream which is suddenly augmented with labels?
• This state might also mean something in the case of a detector with a burn-in period or a window that needs filling, like DDM or ADWIN.

For now we can put this on the backburner, until we have some more use cases where this would be relevant. Ensembles and other kinds of pipelines might be able to make use of it later. Each of these use cases don't really mean the same kind of thing, but it might be useful to have something generic.