Fixing sagemaker get_model() for xgboost

[igorborgest]

Fixing sagemaker get_model() for xgboost

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[tuliocasagrande]

It's nice to remove pickle and leave the deserialization to the user as we have several ways to do it, depending on the model.

I tested a few ones and I'll leave it here for future reference:

1. XGBoost [1]:

   • model.tar.gz contains xgboost-model, which can be loaded with pickle (xgboost.core.Booster)

2. k-Means [2][3]:

   • model.tar.gz contains model_algo-1, which can be loaded with mx.ndarray.load or mx.ndarray.load_frombuffer

3. PCA [4]:

   • model.tar.gz contains model_algo-1, which can be loaded with pickle (dict with keys s, v, mean)

4. SDD [5]:

   • model.tar.gz contains several files (hyperparams.json, model_algo_1-0000.params, model_algo_1-symbol.json)