Speed up the process of users' digit images prediction

[faustineinsun]

Instead of running **/ml-spark-0.0.1-SNAPSHOT-jar-with-dependencies.jar and loading model everytime for each input digit image, it's better for us to find a way to only load model once and then do the prediction for all of the coming digit images multiple times.

For example:

val lrLBFGSModel = LogisticRegressionModel.load(sc, lrLBFGSModelPath) // run only once
val predictions = lrLBFGSModel.predict(testing) // run multiple times