hdrda_cv should return a trained classifier

[ramhiser]

Rather than having to extract the tuning optimal tuning parameters and then train a new classifier, the hdrda_cv function should do this automatically.