added initialize variable to the SKMeans SC class definition

[tremblap]

this seems to work, although with the code below I do not see much of an improvment, so maybe have done something wrong

~ds = FluidDataSet(s).read("/Users/pa/Documents/recherche/projets/flucoma/sources/flucoma-max/misc/rays.json");

(
~ds.dump({
	arg dict;
	defer{
		~fp = FluidPlotter(dict:dict,xmin:-1,ymin:-1).pointSizeScale_(3);
	};
});
)

(
~ls = FluidLabelSet(s);
~skm = FluidSKMeans(s,maxIter: 1000);
)

// 10 inits, graphically, one per second
(
Routine{10.do{
~skm.clear;
~skm.fitPredict(~ds,~ls,{
			~ls.dump({
				arg dict;
				~fp.categories_(dict);
	})
});
	1.wait;
};}.play;
)

// rerun above with other initialisation
~skm.initialize_(1)

[weefuzzy]

I now realise after some prodding that the rays data also presents a difficult case for initialising the means (which is good). I'll put some comparison plots up on the core PR so you can see

[weefuzzy]

KMeans will need updating too, and the possible modes are now 0, 1, 2