perf(gargantua): 渲染按 64 像素分块并每块让出 CPU

原实现一次连续算 4 行（640 像素，单像素最坏 100 步），长时间不让出，单核下输入/合成全卡。改为 64 像素一块，每块后 process.yield_cpu()；每 4 行仍 root.redraw() 保持渐进刷新节奏；事件轮询从每 4 行一次提升到每 64 像素一次。单次占用降到约 1/10，逐像素结果不变（仅调度顺序变化）。

Apps/Gargantua/main.vp

@@ -19,7 +19,8 @@ define DISK_OUT = 11.0
 define CAM_DIST = 22.0
 define CAM_ELEV = 9.0
 define FOVF = 0.55
-define YIELD_ROWS = 4
+define SLICE_PIXELS = 64
+define FLUSH_ROWS = 4
 define V3_POOL_SIZE = 8192
 define V3_POOL_BYTES = V3_POOL_SIZE * 24 + 512
 
@@ -249,6 +250,8 @@ export def _start() -> int:
 
     gT: double = 0.0
     rendered: int = 0
+    rx: int = 0
+    ry: int = 0
 
     evt: window.InputEvent
     while window.is_active(root.win_id):
@@ -263,22 +266,21 @@ export def _start() -> int:
                 canvas.set_fb(fb, new_w, new_h)
 
         if rendered == 0:
-            y: int = 0
-            while y < GH:
-                for x in range(GW):
-                    pixel: uint32 = shade_pixel(x, y, gT)
-                    fb[y * GW + x] = pixel
-                y += 1
-                if y % YIELD_ROWS == 0:
-                    root.redraw()
-                    process.yield_cpu()
-                    result2: int = window.poll_event(&evt)
-                    if result2 != 0:
-                        if evt.type == window.EVENT_TYPE_PING:
-                            window.flush(root.win_id)
-            root.redraw()
-            rendered = 1
-            gT += 50.0
+            # 固定像素预算切片，保证单次占用时间片有上界
+            budget: int = SLICE_PIXELS
+            while budget > 0 and ry < GH:
+                fb[ry * GW + rx] = shade_pixel(rx, ry, gT)
+                rx += 1
+                budget -= 1
+                if rx >= GW:
+                    rx = 0
+                    ry += 1
+                    if ry % FLUSH_ROWS == 0:
+                        root.redraw()
+            if ry >= GH:
+                root.redraw()
+                rendered = 1
+                gT += 50.0
 
         process.yield_cpu()