[Time Series prediction model]Add ARIMAWithSTLDecompostion model

[sneaxiy]

The procedure of the model is:

• Apply STL decomposition to extract the periodic features of the trained data.
• Substract the periodic features from the original trained data, and obtain left_values.
• Use an ARIMA model to fit left_values.
• When prediction, the ARIMA.predict is first called, and the result would be added with the periodic features extracted by the training process.