Improve handling of combining/averaging multiple recordings from same participant

[craddm]

Currently if you average over multiple recordings from a single participant, the way to handle the recording column is unclear, as clearly you don't want (usually) want to average sets of trials separately for each recording!