MultiTask Improvements

[AntonioCarta]

• loss & accuracy are task-agnostic by default. Improves logging in multitask settings.
• classes ids from zero in each exp has been set to False on all the benchmarks. Previously it was True and it caused a lot of surprise and bugs to end users.
• removed benchmark from EvaluationPlugin arguments. The benchmark is taken directly from the current experience. Removed the corresponding warning.

[coveralls]

Pull Request Test Coverage Report for Build 2695872234

• 70 of 90 (77.78%) changed or added relevant lines in 19 files are covered.
• 46 unchanged lines in 5 files lost coverage.
• Overall coverage increased (+0.02%) to 72.257%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
avalanche/benchmarks/classic/ccifar10.py	0	1	0.0%
avalanche/benchmarks/classic/ccifar100.py	0	1	0.0%
avalanche/benchmarks/classic/ccub200.py	0	1	0.0%
avalanche/benchmarks/classic/cimagenet.py	0	1	0.0%
avalanche/benchmarks/classic/cinaturalist.py	0	1	0.0%
avalanche/benchmarks/classic/comniglot.py	0	1	0.0%
avalanche/benchmarks/classic/ctiny_imagenet.py	0	1	0.0%
avalanche/benchmarks/classic/ex_model.py	1	2	50.0%
avalanche/evaluation/metrics/forgetting_bwt.py	1	3	33.33%
avalanche/evaluation/metrics/forward_transfer.py	1	3	33.33%

Files with Coverage Reduction	New Missed Lines	%
avalanche/evaluation/metrics/accuracy.py	1	80.31%
avalanche/evaluation/metrics/loss.py	1	93.26%
avalanche/evaluation/metric_utils.py	1	43.75%
avalanche/evaluation/metric_definitions.py	5	74.88%
avalanche/benchmarks/utils/data_loader.py	38	81.28%

Totals
Change from base Build 2634452838:	0.02%
Covered Lines:	12765
Relevant Lines:	17666

---

💛 - Coveralls