Store cluster membership

[xmacex]

Ok amends the plot_data.json file with k-means clustering. The status quo master branch calculates a clustering to select hotspot images for human consuption on the web UI. With this both the membership for the centroids as centroids/members, as well as the labels are retained as the fifth list element for each of the images under positions in the output file.

So from

{
    "centroids" : [
	{
            "label" : "Cluster 1",
            "img" : "img_021_Brazil_pt_global_warming.jpg"
	},
	{
            "label" : "Cluster 2",
            "img" : "img_125_Brazil_pt_global_warming.jpg",
	}
    ],
   "positions" : [
      ["img_124_Brazil_pt_climate_change_mitigation", -785, 740, 32, 24],
      ["img_078_Brazil_pt_climate_change", -456, -94, 32, 22],
      ["img_022_Brazil_pt_global_warming", -380, 349, 32, 26],
      ["img_219_Brazil_pt_climate_change_adaptation", -735, 746, 32, 24]
    ]
}

To

{
    "centroids" : [
	{
            "label" : "Cluster 1",
            "img" : "img_021_Brazil_pt_global_warming.jpg",
            "members" : [89, 124, 297, 312, 440]
	},
	{
            "label" : "Cluster 2",
            "img" : "img_125_Brazil_pt_global_warming.jpg",
            "members" : [13, 40, 51, 55, 81, 83, 109, 191]
	}
    ],
   "positions" : [
      ["img_124_Brazil_pt_climate_change_mitigation", -785, 740, 32, 24, 1],
      ["img_078_Brazil_pt_climate_change", -456, -94, 32, 22, 2],
      ["img_022_Brazil_pt_global_warming", -380, 349, 32, 26, 2],
      ["img_219_Brazil_pt_climate_change_adaptation", -735, 746, 32, 24, 1]
    ]
}

You get the idea, I hope.

Here's a colouring, with k=20, produced outside of the web UI.

The solution needs validation. From let's say engineering and software design point and it would be nice to confidently say this PR doesn't break anything, e.g. the JSON file parsing by the visualization data load or something. From design or research tooling point of view it is worth asking whether making available a discrete cluster membership is a good idea for working forward from the plot_data.json, because it is a certain violence done on the more topological, more continuous space which we are working with here.

Closes #74.