Add preprocessing for data with missing values in test set

[pplonski]

There can be a situation where new data has missing values in columns that have all values in train dataset. In such case please:

• warn the user about this
• fill missing values with median