improvements to ensclus

This branch provides the following improvements to CST_EnsClustering.R

1. the closest_member output field used to be an index in a combined dimension of all cluster_dim. So if cluster_dim=c("member, "sdate") with dimensionality c(member=15, sdate=6) for example, the result result$closest_member would be an array with indices between 1 and 15*6=90. Now the fields result$closest_member$member and result$closest_member$sdate (or others, depending on the field names given in cluster_dim) will provide the correct indices in those two dimensions for each representative cluster member.
2. introduced a new argument time_dim which allows to select the dimension(s) (can be a list of strings) on which to apply the statistic. Before the statistic was applied always to the 4th dimension, now the order of dimensions is irrelevant. If not specified the time dimension is found automatically, searching in order between c("ftime", "sdate", "time")