Remove dataset-dependent args in CCA pipelines

[sylvchev]

trafficstars

Existing CCA-based pipelines (CCA, TRCA, MsetCCA, see https://github.com/NeuroTechX/moabb/blob/79cbb71c1ed116d7e5bc9368311aff6f3e2b5f19/moabb/pipelines/classification.py#L14) need information about sampling frequency, stimulation frequencies, and/or epoch interval. This shouldn't be the case as it makes the pipelines dataset-dependant.

A possible solution is to change the code from those sklearn estimators to take MNE epochs rather than ndarray. It should be then possible to access directly to the sfreq and interval from the MNE object. The stimulation frequencies could be directly infered from the labels.

This issue requires to rewrite part of the code those CCA pipeline but does not requires any specific knowledge on the CCA or the methods themselves.