WIP: regularization as callbacks

[Xylambda]

Added

• Add torch.ditribution example, with code taken from Romain Strock.
• Add predict method to Trainer. ENH: add predict method to the trainer #38
• Add functions to freeze and unfreeze model. ENH: freeze and unfreeze model #43
• Add function to transform dataset into time series dataset.

Fixed

• Metrics are now moved to the execution device BUG: metrics are not being moved to execution device #41.
• Log level is now used in the Trainer. BUG: log_level is not being used in the Trainer #40
• LearningRateScheduler now does not crash in first epoch when on_train is False. BUG: LearningRateScheduler crashes on first epoch if 'on_train' is False #36

Changed

• Make regularization part of the callbacks system. ENH, DISC: regularization as callbacks #37
• Divide utils into three submodules: convenience,preprocessing and data.
• Update requirements to avoid conflicts.
• Update some tests.

Removed

• Remove old regularization module and all related code.

Closes: #36 #37 #38 #40 #41 #43