Refactor the CLEAR training example

[linzhiqiu]

1. We add a metric to our benchmark
2. We add an example on how to use the linear features of CLEAR to train and evaluate

[coveralls]

Pull Request Test Coverage Report for Build 2292639743

• 9 of 28 (32.14%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.06%) to 70.569%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
avalanche/benchmarks/classic/clear.py	9	28	32.14%

Totals
Change from base Build 2277246645:	-0.06%
Covered Lines:	12037
Relevant Lines:	17057

---

💛 - Coveralls

[AntonioCarta]

Thanks!

[linzhiqiu]

@AntonioCarta I am confused by the failed unit test -- it says internal server error? I wonder what we could do to fix it?

[AntonioCarta]

This is a problem with some of the tests that download the dataset. I will merge this anyway since it's a momentary error.