v0.1.12: downsampling; total path length; various improvements and bug fixes

[huangziwei]

New

Post-processing: downsample skeleton

The skeleton output from skeletonize() preserves a lot of details, which is good in some cases (such as finding contacts between cells), but bad for others (too many nodes to build a biophysical model -> inefficient).

This PR added skeleton downsampling (post.downsample()) to reduce the number of nodes while mostly keeping the total volume and the radii distribution stable (that is, only nodes with similar radii got merged, mainly depending on the rtol parameters).

Diagnostics: total path length

On top of dx.volume(bbox), we now also have bounding box-aware dx.total_path_length(bbox).

Bug Fixes

• io.save_contact_sites_npz() and io.load_contact_sites_npz() should now save and load the latest ContactSites object with all the new fields (bbox*) correctly.

Improvements

• pair.map_contact_sites() exits early with an empty ContactSites when contact_seeds is empty.
• skel_cmap in plot.projection() aka sk.plot2d() now can take dictionary as input, which can override the default colormap. e.g. skel_camp={"soma": "black", "dendrite":"green"}.