Add IIRC scenario (?)

[vlomonaco]

We should check how to integrate this scenario in our framework: https://chandar-lab.github.io/IIRC
The difference its mostly in the evaluation part, where a sample have multiple correct labels.

[AntonioCarta]

My guess from a quick look is that you only need a different Accuracy metric, the rest should work as is. It could be a good test to check if avalanche works with different metrics and ensure that the accuracy is not hardcoded anywhere (as it was in the old eval plugin).

[ggraffieti]

I agree with @AntonioCarta, I also quick-checked the site and seems that we can smoothly integrate the scenario just using an ad hoc accuracy metric.
If I understood well, seems every data point has 2 labels, general and fine, and you have to guess one of the two given the other one. I don't understand how a strategy can use the given label, but in any case the how-to will be delegated to the particular strategy.