Add tract density image computation to evaluation framework

[ebrahimebrahim]

Add a step to the eval framework that computes a tract-visitation-count (aka tract density map) image from the tractography generated by tractseg.

DTIProcessToolkit has code that computes these but it should not be too hard to do it ourselves in python.

The usefulness of this is:

• We can threshold the tract density image to determine which tracts to remove from tractseg bundles to clean them up. This makes our evaluation less sensitive to tractseg. (Use tract density image to clean TractSeg bundles #29)
• We can create a new evaluation metric that is based on comparing tract density images. It should be faster to compute than tract distance, while still capturing some better sort of tract alignment score than tract dice would. (Create evaluation metric based on tract density image #28)
• We can add together tract density images over a set of bundles to see how much of the white matter space they cover. This can help us to select a set of bundles to use for evaluation. (Select set of bundles to use for evaluation #27)