Update the EnsembleDetector with multiple voting system

[anagainaru]

The EnsembleDetector existed before but was not usable, it also only accepted a voting majority as the strategy to decide when to fire. This PR extends it so it can accept two more strategies (any reacts to the first detector that fires, unanimous requires full agreement, majority sits in between).

This PR also fixes the lack of configuration options that made the EnsembleDetector not usable. After the current PR toml files can include the following options:

detector_name = "EnsembleDetector"
ensemble_detectors = ["ADWINDetector", "KSWINDetector", "PageHinkleyDetector"]
ensemble_voting = "majority"

EnsembleDetector builds each name in ensemble_detectors from the same [drift_detection] block (so a detector type can appear at most once). An unknown voting name or an empty detector list raises ValueError.

SLAC model harness (from PR #83) with EWC and KSWin fires 4 times:

with PageHinkley fires 2 times:

And with both (and any strategy) fires 6 times:

[andrewfayres]

One small nitpick

[andrewfayres]

The Config still should still hold the detector name in it. We should continue to derive detector name from the config instead of accepting it separately. It's slightly less error prone and a bit easier to use.

Changing it to this means every caller will do:
detector_name = cfg.drift_detection.detector_name
_build_detector(detector_name, cfg)

instead of:
_build_detector(cfg)

[anagainaru]

The load_drift_detection function is still using the detector name set in config. This change is necessary (i.e. splitting the load function into _build_detector that takes a given detector) to cover the EnsembleDetector. For the EnsembleDetector the config is like this:

detector_name = "EnsembleDetector"
ensemble_detectors = ["ADWINDetector", "KSWINDetector", "PageHinkleyDetector"]
ensemble_voting = "majority"

which means we are calling _build_detector for each detector in the list.

[andrewfayres]

Okay so for the ensemble detector, the name that we're passing to buld_detector will differ from the name in the config? That's what it looks like based on the below code snippet. Odd behavior but it's an internal only method so I suppose it's fine.

detectors=[_build_detector(name, cfg) for name in sub_names]

[andrewfayres]

Okay so for the ensemble detector, the name that we're passing to buld_detector will differ from the name in the config? That's what it looks like based on the below code snippet. Odd behavior but it's an internal only method so I suppose it's fine.

detectors=[_build_detector(name, cfg) for name in sub_names]

[anagainaru]

@andrewfayres we basically check if the name is EnsembleDetector and if yes we go over the list of detectors in the ensemble and build one for each entry. For any other name we just build the given detector.

    detector_name = cfg.drift_detection.detector_name
    if detector_name != "EnsembleDetector":
        return _build_detector(detector_name, cfg)

The other option was to allow the name of the detector to be a list, in which case we would still need to handle it differently depending on if a string or a list was given. Keeping a separate EnsembleDetector allows me to have configuration options for it (e.g. the voting method).