forests get Type, so Predict can be generic.

[glycerine]

Early draft, but almost done:

• draft PR posted for feedback (wait to merge).
• still todo: read back in the ForestType from the serialized text file.
• test that Predict() computes the same answers as applyforrest

[glycerine]

Serialization part is done.

Needs review and in particular the decoding of options into the ballot box type needs comparison with the applyforest.go logic.

[glycerine]

I made a stand alone command,cfpred, that exercises the Predict() functionality.

Initial comparison to applyforest on the iris and foresfires data set, under classifier and -gbt regression showed identical predictions.

[glycerine]

Okay. This is ready to review, and if acceptable, merge.

• Test of Predict() in place in pred_test.go.
• To reduce risk, I've kept the newlogic in cfpred and left applyforest the same. A bit conservative, but I don't want to mess up current users/uses of applyforest. cfpred should be the way forward for new development, and should eventually replace applyforest once it matures.