This PR adds a modified version of the wavefront skeletonization in which the wave propagates exactly a user-defined distance along the mesh instead of hopping from vertex to vertex.

Advantages

• works much better on coarse meshes or meshes with inhomogenous details

Disadvantages

• computationally more expensive (this still needs testing; I also haven't spent any time optimising)
• because this doesn't operate directly on the vertices (but rather the edges between them) there is no 1:1 correspondence between mesh vertices and skeleton nodes (i.e. no mesh_map); we could generate an approximation but that would need to be both ways (vertex -> node and node -> vertex)

Here is a quick and dirty example using one of the navis example neurons:

>>> import navis
>>> import skeletor as sk
>>> mesh = sk.pre.fix_mesh(navis.example_neurons(1, kind='mesh').trimesh, remove_disconnected=20)
>>> # New method moving exactly 50 units at each step
>>> s = sk.skeletonize.by_wavefront_exact(mesh, step_size=50)

Arrows highlight areas where the exact method (yellow) does better than the original (red) method.