k-means initialization procedures

[corybrunson]

There exist a variety of sampling procedures, intended for use with k-means clustering, that are similar in spirit to those in development here but random rather than deterministic: see Ostrovsky &al (2012) and Arthur & Vassilvitskii (2006). Some of these are implemented, but to my knowledge only in k-means–specific packages like flexclust. In the medium term, it may be worthwhile to provide them here, to make multi-package analysis pipelines more flexible.