I suppose the pathfinding is some kind of finding a way
from tile xy to another tile xy.
And the village having more than 10 tiles in x and y direction.

Evolving a pathfinding without any "precoding" of how
it should work will use huge amounts of "evolution-time"


The simulation would take forever to bring meaningful results
(especially if the fittnes function is just live or die)
-> the first hundret thousand of generations would simply die, trying
to find any meaningful path to move on.

"you will end up with lots of dead viligers in the beginning"
exaclty that. But how are you supposed to select the fittest then, if
everyone is making sonesense. Evolution needs not just mutation, but also selection.
But selection means that there is are at least two parts, a good and a less good (bad).
If everyone is bad, there is not much to select.


GA can do lots of things, but the simulation must allow
the GA to solve that in a reasonable time.

Evolution is a type of search algorythm, where selection is the heuristic (evaluating
the quality of the direction) and mutation is the progression-step in the search.

-> roughly translated to A* : the "selection" is the heuristic
determining if the next step is a lower eucledian distance than the previous,
and "mutation" is the step going to another node/tile.
A* can remember past moves, but evolution forgets them at each new step.
So you need at least a heuristic to keep a senseful direction.