Update OpenCV DNN Detection component to optionally use feed-forward confidence values

[jrobble]

Keyword Tagging and Azure Translation just copy the confidence from the feed-forward track to the output track. Same goes for the detections within the track. OCV DNN is different because it can be used as a generic classifier in addition to a color characterizer. There can be cases when we want to have the generated tracks and detections use the OCV DNN confidences and cases when we don't. Modify that component to look at the value of OUTPUT_MERGE_WITH_PREVIOUS_TASK to figure out what to do.