fix: 掩膜降采样计算 — 杜绝 1m 分辨率下 10.5M Point 对象 OOM/假死

真凶定位: prepare_shared_context 第 2982-2988 行
  grid_xx.ravel() → 10.5M 坐标点
  [Point(x,y) for ...] → 创建 10.5M 个 shapely Point (~2 GB)
  within(boundary.unary_union) → 10.5M 次几何测试 O(N×M)
  → 此处假死/卡死/内存爆炸，Kriging 从未开始执行

修复: 掩膜降采样到 ~50K 点 (约 100m 等效分辨率) 计算,
  结果通过 np.kron 升采样回原始分辨率。
  掩膜是布尔值，精度损失可忽略；
  np.kron 的 C 向量化升采样 <1 秒。

src/postprocessing/map.py

@@ -2978,14 +2978,31 @@ class ContentMapper:
 
         print(f"[共享上下文] 网格: {nx}×{ny} = {nx*ny} 点")
 
-        # ⑥ 水域掩膜布尔矩阵（只此一次）
+        # ⑥ 水域掩膜布尔矩阵（降采样到 ~100m 分辨率计算，避免 1m 下千万级 Point 对象 OOM）
         mask = None
         if boundary_gdf is not None:
-            mask_pts = np.column_stack((grid_xx.ravel(), grid_yy.ravel()))
+            _MASK_TARGET_POINTS = 50000  # 掩膜降采样目标点数
+            _mask_step = max(1, int(np.sqrt(grid_xx.size / _MASK_TARGET_POINTS)))
+            if _mask_step > 1:
+                mask_xx = grid_xx[::_mask_step, ::_mask_step]
+                mask_yy = grid_yy[::_mask_step, ::_mask_step]
+                print(f"[共享上下文] 掩膜降采样 {_mask_step}× → "
+                      f"{mask_xx.shape[1]}×{mask_xx.shape[0]} = {mask_xx.size:,} 点")
+            else:
+                mask_xx, mask_yy = grid_xx, grid_yy
+
+            mask_pts = np.column_stack((mask_xx.ravel(), mask_yy.ravel()))
             mask_gdf = gpd.GeoDataFrame(
                 geometry=[Point(x, y) for x, y in mask_pts], crs=self.output_crs
             )
-            mask = mask_gdf.within(boundary_gdf.unary_union).values.reshape(grid_xx.shape)
+            mask_lowres = mask_gdf.within(boundary_gdf.unary_union).values.reshape(mask_xx.shape)
+            # 升采样回原始分辨率（最近邻，掩膜是布尔值）
+            if _mask_step > 1:
+                mask = np.kron(mask_lowres, np.ones((_mask_step, _mask_step), dtype=bool))
+                # 裁剪到精确原始尺寸
+                mask = mask[:grid_xx.shape[0], :grid_xx.shape[1]]
+            else:
+                mask = mask_lowres
             print(f"[共享上下文] 水域掩膜: {int(mask.sum())}/{mask.size} 点在水域内")
 
         return (grid_xx, grid_yy, mask, bounds, boundary_gdf)