A* is only usable for a single path, except with a heuristic function
returning always zero, which turns it into Dijkstra's algorithm.
Distribution uses it that way, because it needs to find multiple paths
from the same source as efficiently as possible. Only the other uses
of path search can profit from A*'s superior efficiency.
I feel the extra complexity is not justified. Besides, it slows down
distribution path assembly a bit, which is the only case where
efficiency really matters. Let's stick to Dijkstra's for now.