fix(phase2): llc 并行线程数组改用 list[Thread]

c.Deref 对 i8* 无法推断加载类型时只读 1 字节（旧编译器），th_ptrs 存
Thread* 后 join/is_alive 拿到垃圾句柄立即失败——llc 并行"5 秒误报失败"
的直接原因（ClassArrayTest 2 模块并发必现）。改 list[Thread] 保类型，
get 返回明确类型，不再依赖 Deref 推断。含 [LLC-DBG] 插桩，验证后移除。

App/lib/core/BuildPipeline.vp

@@ -596,9 +596,11 @@ def run_llc_parallel(tasks: bytes, task_count: int, max_threads: int) -> int:
         return 0
     if max_threads <= 0:
         max_threads = 1
-    # Thread 对象指针数组（stdlib，保证线程对象存活到 join）
+    # Thread 对象数组（list 保类型）：c.Deref 对 i8* 无法推断加载类型时
-    th_ptrs: bytes = stdlib.malloc(t.CSizeT(max_threads) * 8)
+    # 只读 1 字节 → 垃圾指针 → join/is_alive 立即失败（ClassArrayTest llc
-    if th_ptrs is None:
+    # "5 秒误报失败"根因）；list[Thread].get 返回类型明确，无推断问题。
+    th_list: list[vthreading.Thread] | t.CPtr = list[vthreading.Thread](_mbuddy, max_threads)
+    if th_list is None:
         VLogger.error("run_llc_parallel: 线程数组分配失败", "PhaseB")
         return 1
     slot_size: t.CSizeT = LlcTask.__sizeof__()
@@ -615,14 +617,12 @@ def run_llc_parallel(tasks: bytes, task_count: int, max_threads: int) -> int:
                 t.CUInt64T(tasks) + t.CSizeT(batch_start + j) * slot_size, t.CPtr))
             th: vthreading.Thread | t.CPtr = vthreading.Thread(_llc_task_worker, t.CPtr(tsk))
             th.start()
-            # u64 槽存 Thread*（转 u64 存储，读取时再转回指针）
+            th_list.set(j, th)
-            c.DerefAs(t.CUInt64T(t.CUInt64T(th_ptrs) + t.CSizeT(j) * 8, t.CPtr), t.CUInt64T(th))
             j += 1
         # 本批 join
         j = 0
         while j < batch_n:
-            th_j: vthreading.Thread | t.CPtr = (vthreading.Thread | t.CPtr)(t.CPtr(c.Deref(
+            th_j: vthreading.Thread | t.CPtr = th_list.get(j)
-                t.CUInt64T(t.CUInt64T(th_ptrs) + t.CSizeT(j) * 8, t.CPtr))))
             #region debug-point LLC7
             stdio.printf("[LLC-DBG] join 前 %d\n", j)
             #endregion
@@ -651,7 +651,6 @@ def run_llc_parallel(tasks: bytes, task_count: int, max_threads: int) -> int:
                 ret = 1
             j += 1
         batch_start += batch_n
-    stdlib.free(th_ptrs)
     return ret