mne prewhiten

[schoffelen]

This is a PR that eventually should address #1473.

[schoffelen]

The first step is to inventorize for ft_inverse_mne, whether the 'prewhiten' option behaves the same as a call to ft_denoise_prewhiten, followed by ft_sourceanalysis

[schoffelen]

Currently, the test function checks whether ft_sourceanalysis (with 'mne'+'prewhiten') results in the same output as a sequential call to ft_denoise_prewhiten/ft_sourceanalysis (with 'mne' + 'prewhiten'/'no-prewhiten').
With prewhitened data in the input (i.e. an eye(nchan) noise cov matrix), AND cfg.mne.prewhiten = 'yes', the results are the same.

With prewhitened data in the input AND cfg.mne.prewhiten = 'no', the results will only be the same IF cfg.mne.lambda is square rooted :(. this is the consequence of the fact that the estimation of the inverse operator in ft_inverse_mne applies a different formula, and is incosistent in its treatment of lambda (squared vs. non-squared). See #1307