Add CALM Benchmark

[vlomonaco]

Paper: https://arxiv.org/pdf/2004.03340.pdf
@AndreaCossu & @AntonioCarta what do you think?

[AntonioCarta]

Once we fix multitask scenarios I think we can start to tackle online learning scenarios. This is a good one to also test slightly different data formats (e.g. labels at each timestep).