Increase PathFinder maxRooms maximum to 64

[laverdet]

Some users have said that increasing this limit would be useful to them. Since any time the pathfinder enters a room it counts against maxRooms, you usually end up with a maximum path shorter than 16 rooms long.

With the existing code the highest we could bump the k_max_rooms constant to was 24, so I went through and audited and updated the data structures in use. This increases memory usage a little, but reduces CPU overhead a little bit too (by about 7%). The pathfinder is almost allocation-free which means that it allocates all the memory it could possibly need upfront and holds on to it for the program's lifetime. The old code would use ~49kb of memory per thread per k_max_rooms, plus about 74kb of other overhead. The new code uses ~84kb of memory per thread per k_max_rooms, plus the same overhead.

Assuming a 4 thread server, before this patch the pathfinder will hold onto ~3.63mb of memory. After this patch it will hold onto 17.27mb of memory. I don't think that number is unreasonable by any measure but I did want to make sure to communicate the exact cost of this change. And like I mentioned it is 7% faster, at least on my system and probably most 64-bit systems.

This also includes the fix for the "Max heap" issue we discussed a couple days ago. It'll automatically adjust the heap size to the correct capacity based on k_max_rooms.

[artch]

Looks good, thanks! It will be merged into the next PTR patch. Could you please rebase it onto ptr branch? Especially because it already contains the fix in e897bda commit.