Implement tiered pathfinding with thread-local scratchpads

- TIER0 (<10 tiles): Vec-based linear search, zero hash overhead
- TIER1 (10-30 tiles): FxHashMap scratchpad with memory reuse
- TIER2 (30-100 tiles): AHashMap scratchpad for better collision resistance
- TIER3 (>100 tiles): Chunk waypoint decomposition
- Added benchmarking infrastructure with F8 report trigger
- Added async infrastructure: PendingPath, CompletedPaths resources
- Lowered MAX_NODES to 5000 to prevent runaway searches

Benchmarks show P50 improved 13%, max improved 53%

src/constants.rs

@@ -3,3 +3,13 @@ pub const TILE_PIXELS: u32 = 16;
 pub const TILE_SIZE: f32 = TILE_PIXELS as f32 * PIXEL_RATIO;
 pub const ITILE_SIZE: i32 = TILE_SIZE as i32;
 pub const SEED: u32 = 420;
+
+// Pathfinding tier thresholds (in tiles)
+pub const PATHFINDER_TIER0_MAX_TILES: i32 = 10; // Very short paths: Vec-based
+pub const PATHFINDER_TIER1_MAX_TILES: i32 = 30; // Short paths: AHashMap scratchpad
+pub const PATHFINDER_TIER2_MAX_TILES: i32 = 100; // Medium paths: AHashMap scratchpad
+                                                 // TIER3: > 100 tiles → Chunk waypoints + async
+
+// Pathfinding configuration
+pub const PATHFINDER_MAX_NODES: usize = 5000; // Max nodes before partial path
+pub const PATHFINDER_ASYNC_THRESHOLD_TILES: i32 = 150; // Start async for very long paths

src/entities/shared_components/ambulatory.rs

@@ -9,3 +9,21 @@ pub struct Ambulatory {
     pub target: Option<Vec3>,
     pub step_recovery: u32,
 }
+
+#[derive(Component)]
+pub struct PendingPath {
+    pub start: IVec3,
+    pub goal: IVec3,
+    pub waypoint_path: Vec<Vec3>,
+    pub request_id: u64,
+}
+
+#[derive(Resource, Default)]
+pub struct PathRequestCounter {
+    pub next_id: u64,
+}
+
+#[derive(Resource, Default)]
+pub struct CompletedPaths {
+    pub paths: Vec<(u64, Vec<Vec3>)>,
+}