Make NeuronTrace coordinate units more clear #358
Description
Neuron trace coordinates typically come in one of two coordinate systems: physical (e.g. microns) or voxel (e.g. coordinates relative to corner of image, in units of voxels). Cloudvolume handles this by giving skeleton objects an attribute called space which is either physical or voxel. This is a good model.
Our NeuronTrace object is a little messy when dealing with these coordinate systems. For example, the dataframe may be either physical or voxel coordinates, depending on if some seemingly unrelated argument rounding is True or false (see here). Let's make this more clear