TimeSeriesKmean: support sample_weight on fit

Hi,

I opened an issue before but closed it later and decided to say it here in "Feature Request."

I was wondering if you could modify the TimeSeriesKmean function such that it can accept weight (as a callable function) in its metric_params for calculating the dtw_barycenteraveraging.

So:
metric_params = {'weights: ', my_function(data_points)}

So, it is a function that gets a set of data points (observations) and based on that calculates a weight vector and returns it. It gives the flexibility to the user to define a weight function and apply it throughout the clustering process.

(In my problem, for instance, I modified the centroid of the FINAL RESULT and see that it works better for me. However, if such modification can be applied throughout the whole clustering process (and just the final result), it might better enhance the final clusters and result.)

Best,
Nima

Start at TimeSeriesKmean.fit and the dtw_barycenter_averaging path to trace how metric_params are used during clustering. Clarify the callable weight interface and how its returned vector should flow through fitting. Done means callable sample weights are supported throughout clustering without changing existing behavior.