Should we already anticipate that importances could be a 2D array? And consequently have an aggregation strategy like 'mean', 'median', 'max' ... _Originally posted by @jpaillard in https://github.com/mind-inria/hidimstat/pull/361#discussion_r2401063634_