The k-means arg attributes, from which the entry-points are derived, do not define a weight column argument, not in

This despite the trainer itself supporting weighted training:

So we can use weighted training from the command line, but not via entry-points, and so not via the ML.NET API, which is an unfortunate and I believe unintended functionality gap.

One possible remediation is to merely duplicate what @yaeldekel did in #221 , except for this class. Another possibly better solution is to introduce another convenience abstract class for weighted but unsupervised algorithms, to avoid that duplication, then change both to use that.