Improved pathfinding

[kelindar]

This pull request includes significant changes to the pathfinding logic in the tile package, focusing on improving performance and simplifying the codebase. The most important changes include replacing the heap-based priority queue with a bucket-based frontier, optimizing the pathfinding state management, and updating the benchmarks and tests to reflect these changes. Most of the changes are thanks to quasilyte/pathing (see deck zero-alloc-pathfinding). Replacing the heap with his priority queue and replacing the visited map with my int-int map simplified implementation and achieved general speedup of 5-40x, making me wonder what I was thinking when implementing this pathfinding a few years ago.

Pathfinding Optimization:

• Replaced the heap-based priority queue with a bucket-based frontier for more efficient pathfinding (path.go, point.go). [1] [2] [3]
• Introduced a pathfinder struct to manage pathfinding state and optimized state acquisition and release using a sync pool (path.go). [1] [2]
• Replaced map-based edge storage with intmap for better performance and reduced memory usage (path.go). [1] [2]

Benchmark and Test Updates:

• Updated benchmarks to reflect the performance improvements and added new benchmark results (path_test.go). [1] [2] [3]
• Modified test cases to accommodate changes in pathfinding logic and added debug output for path visualization (path_test.go). [1] [2]

Dependency Update:

• Added github.com/kelindar/intmap dependency to the go.mod file.