Cumulative Strategy and tests implemented

[AndreaCossu]

The cumulative strategy is created via a Cumulative Plugin. I used ConcatDatasetWithTargets on cumulative_training_sets() to merge multiple datasets.
The plugin override make_train_dataset in TrainingFlow.
I verified that the size of training set increases as the number of batches progresses.
I added tests for existing strategies.
Closes #127

[vlomonaco]

Let me know when this is ready to be reviewed!

[AndreaCossu]

I implemented Cumulative as a strategy which extends Naive. Check it out 🤞
I verified that training size increases up to the total number of patterns in the last task.

[vlomonaco]

are the acc results at the end of an MNIST Training good for each task? :)

[AndreaCossu]

I guess 😁 With 100 epochs and a SimpleMLP():

Computing accuracy on the whole test set
[Evaluation] Task 0: Avg Loss 0.005673463326126432; Avg Acc 0.8305262923240662
[Evaluation] Task 0: Avg Loss 0.005646013585128195; Avg Acc 0.8009523749351501
[Evaluation] Task 0: Avg Loss 0.004320864401375151; Avg Acc 0.8669999837875366
[Evaluation] Task 0: Avg Loss 0.0034416350111937115; Avg Acc 0.9095454812049866
[Evaluation] Task 0: Avg Loss 0.003999979075175434; Avg Acc 0.8945000171661377

[vlomonaco]

Ok merging now