Predict using the distribution of tree means, giving each ensemble member equal weight.

Trees whose routed leaf holds no positive weight have no prediction and are skipped, as in mergeForestEvidence. When every routed leaf is empty the result is ForestPrediction(0.0, 0.0).