AFRICA should work on epoched data

[RomeshA]

Issues are

• Manual GUI does not support displaying trials
• Automatic would be fine but metrics don't compute due to array shape errors

In addition, methods that assume the data are continuous (e.g. fft and autocorrelation used for mains and cardiac artefacts) might not work well on epoched data. Kurtosis and channel correlation may be fine if the data are treated as continuous. The proposed solution would be to basically treat the epoched data as continuous (reshaping to eliminate trial dimension) and skipping mains and cardiac autocorrelation. compute_metrics could skip them, and then the automatic and manual detectors could check if those fields exist the same way as they check whether correlations were computed for modalities.