Add IL2M strategy

Hi!

I'd like to implement IL2M in avalanche, along with the metric that was proposed to show bias toward new classes.
reference paper: https://openaccess.thecvf.com/content_ICCV_2019/papers/Belouadah_IL2M_Class_Incremental_Learning_With_Dual_Memory_ICCV_2019_paper.pdf
code: https://github.com/EdenBelouadah/class-incremental-learning/tree/master/il2m

I don't know if there is already a plan to port this or if it's ok if I start working on that?

Start by reading the linked IL2M paper and reference implementation, then inspect Avalanche's existing strategy and metric extension points. The work is done when IL2M and the proposed new-class bias metric are integrated into Avalanche and usable through its public APIs.

Written by the indexing model from the issue text.