Implement Potential/Flowfield pathfinding

[Barsonax]

Summary

Implement flow and potential field pathfinding. The flowfield should be created by using the potential field.

Analysis

Both:

• Efficient multi agent pathfinding.
• Robust in the sense that if a unit somehow gets pushed from the path it will find its way back without problems.
• The naïve implementation that will calculate the field for the entire map will use up more cpu and memory than necessary. This could be made more efficient by only doing a high level hierarchical astar search to determine which parts of the map are needed. Caching fields will also be more complex as you could have a incomplete field. However this requires a lot more work as no such nodenetwork or algorithm has been implemented yet in pathfindax. For now stick to the simple implementation.
• Can and should be cached based on the targetnode, clearance and agentsize (the startnode does not matter). This will reduce memory usage and increase performance.

PotentiaField:

• Less performant than flowfields since the actual heading at a position has to be calculated by checking all neighbours and picking the one with the highest potential. Still this are just a couple of array lookups so its still quite fast. The difference might even be unnoticeable in many cases.
• Dynamic pathfinding if agents can emit their own potential field that's added to the potential field values when figuring out the heading.

FlowField:

• Highly performant once the field has been calculated. Getting the heading for a particular agent at a particular position is basically an array lookup.
• if you constrain the directions to 248 different directions and 1 for the Vector.Zero then it will only take 1 byte per node to store a flowfield if you use a separate lookup array for these values.
• Basically what you would get if you asked the heading of every node in a potential field and store this in an array.

[Barsonax]

Implemented