Add a new benchmark over many datasets

[toncho11]

It does include many P300 and Motor Imagery datasets. It adapts both the paradigms and the pipelines. And most importantly it runs statistics to show if the results are significant (using moabb).

The initial version can be run already.

Currently it requires the latest version of MOABB from github where: - cache_config parameter is availabe for WithinSessionEvaluation() - this bug is fixed: https://github.com/NeuroTechX/moabb/issues/514