[FEA] Group SVM model parameters by class

[tfeher]

Is your feature request related to a problem? Please describe.
Currently our SVM classifier does not follow the same layout as Sklearn / LibSVM while exposing the model parameters through the following attributes:

• n_support_ we have total number of SVs while Sklearn has number per class
• support_idx_ we store it in ascending order, while Sklearn has it sorted by class then ascending order by idx
• support_ sorted as support_idx_
• dual_coefs_ sorted as support_idx_

These differences have no effect on making predictions with SVM, but it is inconvenient if someone wants to explore the parameters of the fitted model.

Describe the solution you'd like
Make our SVM parameter layout compatible with Sklearn.

Additional context
The layout of the parameters becomes more complex once we have multi-class classification in place. This issue should probably be addressed when we implement multi-class classification.