Add support for cancelling of training

[glebuk]

Once the cancellation mechanism is available, we need to enable cancellation of training for the following learners:

• RegressionCatalog.Trainers.StochasticDualCoordinateAscent
• RegressionCatalog.Trainers.FastTree
• BinaryClassificationCatalog.Trainers.FastTree
• BinaryClassificationCatalog.Trainers.LogisticRegression

This is continuation of #2795

At the very least we should be able to check for cancellation at the beginning of training and at the beginning of each iteration.

Pri2: Also do this for:

• Normalization
• OneHotEncoding